AI Project Logbook

Resource for Students

(Adapted from "IBM EdTech Youth Challenge – Project Logbook")

PROJECT DETAILS
PROJECT NAME: ________________________ (ADD YOUR PROJECT NAME)
SCHOOL NAME: ________________________ (ADD YOUR SCHOOL NAME)

YEAR/CLASS: 10th / 11th / 12th Grade

TEACHER NAME: ________________________ (ADD TEACHER NAME)
TEACHER EMAIL: ________________________ (ADD EMAIL)

TEAM MEMBER NAMES AND GRADES

1. ________________________ (ADD NAME)
2. ________________________ (ADD NAME)
3. ________________________ (ADD NAME)
4. ________________________ (ADD NAME)
5. ________________________ (ADD NAME)
6. ________________________ (ADD NAME)

(Add more rows if there are more members in your team)

SECTION 2: TEAM ROLES & PLANNING

2.1 Team Roles
 Team
Role Role Description Member
Name
Project Coordinates overall project, manages _____________
Lead timeline, ensures milestones are met ___________
Data Collects, organizes, and labels training _____________
Manager data for the AI model ___________
AI Builds and trains the AI model using _____________
Developer selected tool ___________
UI/Design Creates user interface and designs _____________
Lead solution presentation ___________
Testing Conducts user testing and gathers _____________
Lead feedback ___________
Documenta Records meeting minutes, updates _____________
tion Lead logbook, creates video script ___________

2.2 Project Plan

 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
Course Watc
Dat
work, hed
Pre We We 3 a
readin tutori
par ek ek ho Ma
gs on als on
ing 1 1 urs nag
AI ML4A
er
basics ll
Doc Creat
Set up
um ed
team
We We 1 ent share
folder
ek ek ho atio d
on
1 1 ur n docu
Google
Lea ment
Drive
d s
Backgr
Resea
Pro ound
rched
ble readin Pro
We We 3 JEE
m g on ject
ek ek ho prep
Def educat Lea
2 2 urs arati
init ion d
on
ion challe
issues
nges
Resear
Ident
ch
ified
issues We We 4
8
in ek ek ho All
local
school/ 2 2 urs
probl
comm
ems
unity
Team We We 1.5 Pro Voted
meeti ek ek ho ject &
ng to 2 2 urs select
 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
select Lea ed
issue d "revis
ion
chall
enge"
Compl
ete Probl
Sectio em
We We 2
n3 state
ek ek ho All
(Probl ment
2 3 urs
em finali
Definit zed
ion)
Use
Identif
r Liste
y Test
Un We We 2 d 20
target ing
der ek ek ho poten
users Lea
sta 3 3 urs tial
(stude d
ndi users
nts)
ng
Obser
Watc
ve Test
We We 3 hed
studen ing
ek ek ho study
ts Lea
3 3 urs habit
studyi d
s
ng
Intervi Cond
Test
ew We We 4 ucted
ing
with ek ek ho 3
Lea
users 3 4 urs inter
d
(1-3) views
 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
Compl
User
ete
insig
Empat We We 2
hts
hy ek ek ho All
docu
Map & 4 4 urs
ment
Sectio
ed
n4
Brain
Gener
Bra storm
ate 5
inst We We 2 ing
AI
or ek ek ho All sessio
solutio
mi 4 4 urs n
n
ng comp
ideas
leted
Pro Selec
Create We We 1
ject ted
Priorit ek ek ho
Lea best
y Grid 4 4 ur
d idea
Compl Soluti
We We 1
ete on
ek ek ho All
Sectio finali
4 5 ur
n5 zed
UI Creat
Design
We We 2 Des ed
Des user
ek ek ho ign wiref
ign steps
5 5 urs Lea rame
& flow
d s
Compl We We 1 All Desig
ete ek ek ho n
5 5 ur docu
 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
Sectio ment
n6 ation
done
Liste
Dat Identif Dat
d
a y data We We 1.5 a
chem
Col requir ek ek ho Ma
istry
lect ement 5 5 urs nag
quest
ion s er
ions
Gathe
Collect
Dat red
&
We We 5 a 200
prepa
ek ek ho Ma label
re
5 6 urs nag ed
trainin
er exam
g data
ples
Dat Data
Compl
We We 1 a ethics
ete
ek ek ho Ma check
Sectio
6 6 ur nag list
n7
er done
Pro AI
We We 1 Chose
toty Select Dev
ek ek ho ML4A
pin AI tool elo
6 6 ur ll
g per
Train Achie
AI
AI We We 4 ved
Dev
model ek ek ho 92%
elo
with 6 7 urs accur
per
data acy
 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
Work
Build
AI ing
protot We We 6
Dev proto
ype ek ek ho
elo type
applic 7 8 urs
per comp
ation
leted
Proto
Compl type
We We 1
ete docu
ek ek ho All
Sectio ment
8 8 ur
n8 ation
done
Test Conta
Invite We We 1
Test ing cted
users ek ek ho
ing Lea test
to test 8 8 ur
d users
Teste
Condu
Test d
ct We We 4
ing with
testing ek ek ho
Lea 8
sessio 8 9 urs
d stude
ns
nts
Feed
Test
Collect We We 2 back
ing
feedba ek ek ho grid
Lea
ck 9 9 urs comp
d
leted
Make We We 3 AI Fixed
urgent ek ek ho Dev critic
impro 9 9 urs
 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
vemen elo al
ts per issues
Testi
Compl ng
We We 1
ete docu
ek ek ho All
Sectio ment
9 9 ur
n9 ation
done
Doc
3-
Vid um
minu
eo We We 2 ent
Write te
Cre ek ek ho atio
script script
atio 9 9 urs n
writt
n Lea
en
d
Recor
We We 3 ded
Film
ek ek ho All prese
video
10 10 urs ntatio
n
UI
Edit &
We We 3 Des Video
finaliz
ek ek ho ign edite
e
10 10 urs Lea d
video
d
Team
Co
reflect We We 1.5 Final
mpl
ion ek ek ho All discu
etio
meeti 10 10 urs ssion
n
ng
 Pla Ac
Pl Ac Ac
Pla nn tu
an tu tu Res
nn ed al
Pha ne al al pon Note
Task ed Du Du
se d St E sibl s
Sta rat rat
En ar n e
rt io io
d t d
n n
Compl
Refle
ete We We 2
ctions
Sectio ek ek ho All
writt
ns 10- 10 10 urs
en
11
Doc
Final um
logboo We We 1 ent Quali
k ek ek ho atio ty
revie 10 10 ur n check
w Lea
d
Submi
Pro
t on We We 0.5
ject Subm
IBM ek ek ho
Lea itted
platfor 10 10 urs
d
m

2.3 Communications Plan

Will you meet face-to-face, online or a mixture of each to communicate?
Mixture: Weekly in-person meetings on Friday afternoons + daily WhatsApp group
updates + bi-weekly Google Meet calls
How often will you come together to share your progress?

Weekly meetings (1.5 hours every Friday 4:00-5:30 PM)

Daily updates via WhatsApp (quick status updates)
Bi-weekly detailed sync (alternate Wednesdays via Google Meet)
Who will set up online documents and ensure that everyone is contributing?
Documentation Lead will maintain the shared Google Drive folder and ensure all
sections are updated daily
Project Lead will track individual contributions and follow up if needed
What tools will you use for communication?
WhatsApp - Daily quick updates and questions
Google Drive - Shared documents, logbook, data collection sheets
Google Meet - Bi-weekly team video calls for detailed discussions
Google Forms - User feedback surveys and interview responses
Trello - Task tracking and progress visualization

2.4 Team Meeting Minutes

Meeting #1 - Project Kickoff
Date of meeting: [Week 1, Date]
Who attended: [Add team member names]

Who wasn't able to attend: None

Purpose of meeting: Project introduction, team formation, role assignment, initial
brainstorming of local problems
Items discussed:

1. Overview of AI Project Logbook and requirements

2. Assignment of roles based on strengths and interests
3. Discussion of 8 local problems affecting school/community (student revision
struggles, teacher workload, exam anxiety, subject difficulty, etc.)
4. Initial reaction to problems and preliminary voting
Things to do (what, by whom, by when):
1. Complete coursework readings on AI basics → Data Manager → By Week 1 End
2. Create shared Google Drive folder with template documents → Documentation Lead
→ By Week 1 End
3. Research background on selected problem area → All → By Week 2 Start
4. Schedule user interview appointments → Testing Lead → By Week 2 End

Meeting #2 - Problem Definition

Date of meeting: [Week 2, Date]

Who attended: [All team members]

Who wasn't able to attend: None
Purpose of meeting: Finalize problem statement and prepare for user research

Items discussed:
1. Presentation of background research on education challenges
2. Discussion of 8 identified community issues and pros/cons of each
3. Final vote on which problem to focus on: "Many high school students struggle with
revision because they lack access to personalized practice problems and instant
 feedback on their weak topics"
4. Initial thoughts on how AI could help solve this
Things to do (what, by whom, by when):

1. Finalize problem statement document → All → By Week 2 End

2. Identify and contact potential student users for interviews → Testing Lead → By
Week 2 End
3. Create interview question template → Testing Lead → By Week 3 Start
4. Schedule 3 user interview sessions → Testing Lead → By Week 3 End

Meeting #3 - User Research Complete

Date of meeting: [Week 4, Date]

Who attended: [All team members]

Who wasn't able to attend: None
Purpose of meeting: Share user research findings and complete empathy mapping

Items discussed:
1. Summary of 3 user interviews - key insights about student struggles with revision
2. Empathy map review - what students say, think, do, feel about revision challenges
3. Current user workflow - typical steps students take while revising (read notes →
highlight → re-read → test → check marks)
4. Pain points identified: time-consuming, demotivating when they score low, no
personalized guidance
Things to do (what, by whom, by when):

1. Create empathy map poster → Testing Lead → By Week 4 End

2. Document all user insights in Section 4 → All → By Week 4 End
3. Begin brainstorming AI solution ideas → All → By Week 4 End
4. Prepare 5 AI ideas for evaluation → All → By Week 4 End

Meeting #4 - Solution Selection

Date of meeting: [Week 4, Date]
Who attended: [All team members]
Who wasn't able to attend: None

Purpose of meeting: Brainstorm AI ideas and select best solution using Priority Grid
Items discussed:
1. Presentation of 5 AI solution ideas
2. Evaluation of each idea on "value to users" and "ease of creation" dimensions
3. Priority grid analysis - Best idea: "AI-powered revision assistant using image
classification to detect weak topics and recommend personalized practice
problems"
 4. Selection of ML4All as the development tool (beginner-friendly, free, good accuracy)
Things to do (what, by whom, by when):

1. Document all 5 ideas in Section 5 → All → By Week 4 End

2. Create priority grid visualization → Project Lead → By Week 5 Start
3. Begin detailed solution design → UI Design Lead → By Week 5 Start
4. Start collecting training data (chemistry questions) → Data Manager → By Week 5
Start

SECTION 3: PROBLEM DEFINITION

3.1 Important Local Issues Faced by School/Community
1. Student Revision Struggles - Many students find revision ineffective and time-
consuming; they score poorly on practice tests
2. Teacher Workload - Teachers cannot give personalized feedback to every student
due to large class sizes (50+ students)
3. Exam Anxiety - Students become anxious during exams due to lack of confidence in
weak topics
4. Unequal Access to Resources - Students from lower-income families cannot afford
coaching classes or private tutors
5. Subject Difficulty - Chemistry and mathematics are particularly challenging;
students lose motivation when they score low
6. Limited Feedback Loop - Students submit assignments but get feedback after many
days; by then, they've moved on to new topics
7. One-Size-Fits-All Teaching - Teachers follow standard curriculum; no personalized
learning paths based on individual student needs
8. Technology Access - Many students don't have consistent internet or devices for
online learning

3.2 Which Issues Matter to You and Why?

Student Revision Struggles matter most to us because:
It directly affects our daily lives as students
We've personally experienced frustration when revision doesn't lead to good exam
scores
We can observe this problem firsthand in our school - many classmates struggle
during revision
The problem is solvable using AI - we can create an intelligent system to provide
personalized guidance
It has measurable impact - we can track improvement in student test scores before
and after using our solution

The other issues are important, but they require institutional changes (teacher training,
budget allocation, policy changes). Our AI solution can address the student revision
problem at scale without requiring major systemic changes.
3.3 Which Issue Will You Focus On?
Focus Issue: Student Revision Struggles
Specifically: "Many high school students find revision ineffective because they lack
personalized guidance on which topics to focus on, and they don't get instant feedback on
their practice attempts. This leads to poor exam performance and loss of motivation."

3.4 Problem Statement

How can we help high school students find a way to identify their weak topics and get
instant personalized practice questions so that they can improve their exam
performance and build confidence in their studies?

Self-Assessment for Problem Definition

Rating: 3 points

Justification: A local problem which has not been fully solved before is explained in detail
with supporting research. We interviewed students, identified the specific pain points, and
confirmed that while coaching centers exist, they are expensive and not accessible to all
students. Our AI solution is novel in providing instant, free, personalized revision guidance.

SECTION 4: THE USERS

4.1 Who Are the Users and How Are They Affected by the
Problem?
Primary Users: High school students (Grade 9-12) preparing for board exams and JEE/NEET
entrance exams
How they are affected:

Spend 3-4 hours daily on revision but see minimal improvement in scores
Feel anxious during exams because they're uncertain about weak topics
Cannot afford coaching classes (costs ₹50,000-150,000/year)
Don't receive immediate feedback on their practice attempts (teachers give feedback
after 5-7 days)
Use generic study materials that don't match their specific weak areas
Lose motivation after scoring poorly on practice tests without understanding why
they failed
Secondary Users:
Teachers (want to monitor which topics students are struggling with)
Parents (want to track their child's progress without constant tuition bills)
4.2 What Have You Actually Observed About the Users
and How the Problem Affects Them?
Observations from 1-2 hours of observing 5 students during revision:
1. Students start with enthusiasm but become frustrated after 30-45 minutes when
they can't answer practice questions
2. They don't have a strategy - they randomly pick topics to revise instead of
prioritizing weak areas
3. They re-read notes repeatedly instead of actively practicing, which is passive and
ineffective
4. When they score low on practice tests, they feel demotivated and stop studying;
they don't analyze which questions they got wrong
5. They ask teachers for help but get vague advice like "study harder" instead of
specific guidance on which topics to focus on
6. They switch between multiple resources (YouTube, Vedantu, textbooks) but don't
have a coherent learning plan
7. Marking schemes on tests don't help them understand why they lost marks - they
need detailed explanations
8. Time management is poor - they spend 2 hours on easy topics and 20 minutes on
hard topics, when it should be reversed
9. They study in isolation without knowing if they're better or worse than peers
(competition would motivate them)
10. By 2 weeks before exams, they feel overwhelmed because they haven't adequately
covered weak topics

4.3 Interview Questions and User Responses

Interview #1 - Student A (10th Grade, Chemistry)
Q1: Can you walk me through your revision process?
A: I read my notes, highlight important points, then try practice problems. But I waste a
lot of time reading because I don't know which topics need more focus.

Q2: How do you currently identify which topics are your weak areas?
A: I guess after I score poorly on a test. But by then, it's too late. I've already wasted
time on topics I'm good at.
Q3: What would be most helpful for you during revision?

A: A system that tells me exactly which topics I'm weak in, gives me practice problems
on only those topics, and explains why I got questions wrong. I wish I got instant
feedback instead of waiting a week for my teacher to check my assignment.
Q4: How much would you be willing to pay for such a solution?
A: Nothing. I'm already spending money on tuition. But if it's free on a phone app, I'd
definitely use it.

Q5: What frustrates you most about current revision methods?

 A: Time-wasting and lack of personalization. I'm not a machine - I can't study
everything the same way.

Interview #2 - Student B (11th Grade, Mathematics)

Q1: Can you walk me through your revision process?
A: I watch video tutorials first, then attempt problems. But I get stuck on hard problems
and don't know how to proceed. I wait for my tutor's next class to ask, which is
sometimes 4-5 days later.
Q2: How do you identify weak areas?

A: My tutor tells me after mock tests. But I would prefer instant feedback so I can
immediately work on those topics while the concepts are fresh in my mind.
Q3: What would be most helpful?
A: An app that shows me which topics to practice based on my previous mistakes.
Something that tracks my progress over time and tells me when I'm finally ready to
move to the next topic.

Q4: Would you use an AI tutor?

A: Yes! But it has to explain answers clearly. Just showing "correct" or "wrong" is not
enough.
Q5: How much time do you spend on revision daily?

A: 2-3 hours on weekends, but I don't feel productive. I often study the same topics
repeatedly instead of focusing on new weak areas.

Interview #3 - Student C (12th Grade, preparing for JEE)

Q1: What's your biggest challenge with revision?
A: Choosing which chapters to revise. With 40+ chapters across 3 subjects, I need a
smart system to prioritize. Currently, I just follow my tutor's schedule blindly, but I don't
know if that matches MY weak areas.
Q2: Do you get enough practice opportunities?

A: Yes, but most practice papers are generic. I wish I got practice questions tailored to
MY specific weak topics, not just standard mock tests.
Q3: How much does coaching cost for you?
A: ₹80,000/year. But I still need additional help for chapters I'm weak in. An AI system
could supplement this.
Q4: What would make you more confident before the exam?

A: Clear data showing my improvement over time. Seeing graphs of my accuracy

improving week by week would be motivating.
Q5: Would an AI-powered tool help you?
A: Absolutely. But it needs to be faster than my tutor. I can't wait days for feedback. I
need instant results so I can keep studying while concepts are fresh.

4.4 Empathy Map

What Our Users Are Saying
"I don't know which topics to focus on"
"I'm wasting time studying topics I'm already good at"
"By the time I get feedback from my teacher, I've forgotten what I did wrong"
"I can't afford expensive coaching classes"
"I'm studying so hard but my marks aren't improving"
"I need help, but my tutor is only available 3 hours/week"
"I'm anxious about the exam because I don't know if I'm ready"

What Our Users Are Thinking

"Am I studying the right topics?"
"Will I pass this exam?"
"How do I compare to other students?"
"Why am I not improving despite studying so hard?"
"There must be a better way to study efficiently"
"I'm running out of time"
"Will a tutor help me, or is it too expensive?"

What Our Users Are Doing

Spending 2-4 hours daily on revision
Reading notes repeatedly (passive)
Attempting random practice problems
Waiting for weekly teacher feedback
Attending coaching classes (if affordable)
Watching YouTube tutorials without structured learning
Taking mock tests but not analyzing mistakes thoroughly
Switching between multiple resources
Asking peers "what are you studying?" instead of having a plan
Cramming a few days before exams

How Our Users Feel

Frustrated - spending time but not seeing results
Anxious - uncertain if they're prepared enough
Overwhelmed - too much syllabus, not enough time
Demotivated - low practice test scores make them lose confidence
Confused - don't know which topics need more focus
Impatient - want instant feedback instead of waiting days
Hopeful - believe technology could help them study better
Stressed - especially as exam date approaches
4.5 Current User Steps and Difficulties
Current Revision Process (What Students Do Now)
1. Open study materials (textbook or online notes)
Difficulty: Materials are generic; not personalized to their weak topics
2. Passively read or re-read notes
Difficulty: This is ineffective for long-term retention; students waste time re-
reading topics they already know
3. Highlight important points (or take screenshots)
Difficulty: Too much time spent on highlighting instead of active learning
4. Attempt practice problems (randomly selected)
Difficulty: No guidance on which problems to attempt first; strong students
waste time on easy problems
5. Check answers (use answer key or wait for teacher feedback)
Difficulty: If waiting for teacher, feedback comes after 5-7 days; students lose
momentum; if using answer keys, they get "correct/incorrect" but not detailed
explanations
6. Identify mistakes (if they bother to do this step)
Difficulty: Students don't know WHY they got it wrong; they don't extract lessons
learned
7. Re-attempt similar problems (optional, most students skip this)
Difficulty: Without targeted practice, students don't improve; they make same
mistakes on exams
8. Study next topic (without assessing readiness for current topic)
Difficulty: Students move forward without mastering previous topics; gaps
accumulate
9. Take mock tests (once a week or less)
Difficulty: Mock tests show they're weak in many topics but don't tell them HOW
to improve or WHERE to start
10. Reach exam day with uncertainty
Difficulty: Students don't feel confident; they haven't had a systematic approach
to covering all topics

4.6 User Problem Statement

High school students are experiencing issues with ineffective revision today because
of:
1. Lack of personalization - They use generic materials that don't match their specific
weak topics
2. Slow feedback loop - They get feedback 5-7 days after their attempts instead of
instantly
3. No clear guidance - They don't know which topics to focus on or in what order
4. High cost of tutoring - Private tutors (₹500-1000/hour) are not accessible to all
students
5. Time inefficiency - They waste time on topics they're already good at instead of
prioritizing weak areas
6. Lack of motivation - Without seeing clear progress, they lose confidence and
motivation
 7. Information overload - With 40+ chapters, they feel overwhelmed and don't have a
strategy

Self-Assessment for The Users

Rating: 3 points
Justification: Understanding of the user group is evidenced by completion of all steps in this
section and thorough investigation. We conducted 3 in-depth interviews, observed students
during revision, created a detailed empathy map, documented the current 10-step revision
process with specific difficulties, and identified 7 root causes of the problem. Our
understanding goes beyond surface-level observations; we've analyzed the emotional,
cognitive, and practical dimensions of the user challenge.

SECTION 5: BRAINSTORMING
5.1 Five AI Solutions
AI Idea #1: Image-Based Topic Recognition & Personalized Practice
Concept: Students take a photo of a question from their exam (or textbook). AI recognizes
the topic (e.g., "Organic Chemistry - Benzene Reactions"). AI then recommends 3-5 practice
problems specifically on that topic, generates explanations for each, and tracks which
topics the student photographs most (indicating weak areas). Over time, AI learns the
student's weak topics and proactively recommends revision on those areas.

How AI helps: Image classification + topic database + personalization engine

User benefit: Instant topic identification + targeted practice + focuses on weak areas

AI Idea #2: Quiz Bot with Adaptive Difficulty

Concept: AI chatbot that conducts interactive quizzes. Student starts with easy questions. If
they answer correctly, difficulty increases. If they answer wrong 2x, bot explains the
concept and gives an easier version of the same question. Bot maintains a score and shows
progress graph. Students can ask "quiz me on chemistry" or "quiz me on weak topics from
my last test."
How AI helps: Natural language processing + adaptive difficulty algorithm + performance
tracking

User benefit: Interactive learning + instant feedback + adaptively builds confidence

AI Idea #3: Mistake Analysis & Concept Explanation

Concept: Student uploads a photo of their practice test. AI reads the question and their
answer. AI identifies the specific concept that was missed (e.g., "Error: Confusion between
SN1 and SN2 mechanisms"). AI generates a detailed explanation with 3-4 worked examples
showing the correct approach. AI recommends 5 similar practice problems so the student
can re-attempt and confirm learning.
How AI helps: Image recognition + OCR + knowledge base + explanation generation
User benefit: Detailed error analysis + concept clarity + targeted re-practice

AI Idea #4: Time-Bound Personalized Study Plan

Concept: Student inputs their exam date, current weak topics, and available study hours
per day. AI generates a day-by-day study plan (e.g., "Monday: Revise Organic Chemistry -
spend 2 hours on Benzene reactions because you scored 20% on this topic last test"). AI
adjusts the plan based on quiz scores. Sends daily reminders and motivational messages.
How AI helps: Scheduling algorithm + personalization + data-driven optimization

User benefit: Clear roadmap + no guesswork + confidence from having a plan

AI Idea #5: Social Comparison & Progress Dashboard

Concept: Students join a class group (e.g., "XYZ School 10-B Chemistry"). AI maintains an
anonymous leaderboard showing each student's average accuracy, topics mastered, and
improvement rate. Student can see they're in the 60th percentile on "Organic Chemistry"
and get a recommendation to focus there. AI sends motivational messages like "You
improved 5% this week - keep it up!" and celebrates milestones.
How AI helps: Ranking algorithm + progress tracking + gamification
User benefit: Motivation through social comparison + clear progress visualization +
healthy competition

5.2 Priority Grid

Evaluation Criteria:
Y-axis (Value to Users): How much will this help students improve exam scores and
reduce stress?
X-axis (Ease of Creation): How easily can we build this with available tools (ML4All,
free APIs) in 5-8 weeks?

Grid Placement:
HIGH VALUE + EASY TO CREATE (TOP RIGHT) - Best Choice
Idea #1: Image-Based Topic Recognition & Personalized Practice
Value: 9/10 (directly solves the personalization + weak topic identification
problem)
Ease: 8/10 (image classification is ML4All's core strength; topic database is
simple lookup table)

HIGH VALUE + HARD TO CREATE (TOP LEFT)

Idea #3: Mistake Analysis & Concept Explanation
Value: 9/10 (very helpful - explains concepts)
 Ease: 4/10 (requires advanced OCR + knowledge base + explanation generation;
too complex for 8 weeks)
Idea #4: Time-Bound Personalized Study Plan
Value: 8/10 (helpful - gives clear roadmap)
Ease: 5/10 (requires complex scheduling algorithm + integration with
databases)
LOW VALUE + EASY TO CREATE (BOTTOM RIGHT)

Idea #2: Quiz Bot with Adaptive Difficulty

Value: 6/10 (helpful but generic - not uniquely suited to our users' core pain
point)
Ease: 8/10 (many chatbot builders available; relatively straightforward)
LOW VALUE + HARD TO CREATE (BOTTOM LEFT)
Idea #5: Social Comparison & Progress Dashboard
Value: 5/10 (gamification helps but not core to solving revision ineffectiveness)
Ease: 4/10 (requires backend database + user authentication + leaderboard
system)

5.3 Selected Solution

WINNING IDEA: Image-Based Topic Recognition & Personalized Practice

Brief Summary
Our solution allows students to photograph any chemistry question (from textbook, mock
test, or exam) and instantly get:

1. Topic Identification - AI recognizes which topic the question belongs to (e.g.,

"Organic Chemistry - Benzene Reactions")
2. Related Practice Problems - 3-5 similar problems on the same topic
3. Concept Explanation - A clear explanation of the key concept tested
4. Progress Tracking - System learns which topics the student photographs most often
and recommends focused revision on those topics
5. Personalized Guidance - As the student uses the tool over weeks, AI builds a profile
of weak topics and creates a revision priority list

Why This Solution?

Directly solves the core problem - Students identify weak topics faster than waiting
for teachers/tutors
Instant feedback - Photos processed in <30 seconds; students get guidance
immediately
Personalized - System adapts to each student's specific weaknesses, not generic
Free and accessible - Works on any smartphone with a camera
Practical for JEE/Board exams - Can be used during actual study sessions with real
problems
Achievable in 8 weeks - Uses ML4All (beginner-friendly tool) for image
classification
Tool Selected: ML4All
Why ML4All?
Free, cloud-based platform (no coding required)
Built for beginners and non-technical users
Can train image classification models easily
Provides API for integration into simple mobile app
Supports image uploads and instant predictions
Perfect for high school project scope

Self-Assessment for Brainstorming

Rating: 3 points
Justification: A brainstorming session was conducted using creative and critical thinking.
We generated 5 diverse AI ideas, evaluated them systematically using a priority grid
(considering both user value and creation feasibility), and selected a compelling solution
with strong supporting arguments. The selection was data-driven (based on priority grid
analysis) and well-justified based on user interviews and feasibility analysis.

SECTION 6: DESIGN
6.1 New User Steps with AI Solution
When students use our Image-Based Topic Recognition & Personalized Practice system:
1. Student opens the mobile app (or web interface) on their phone while studying
2. Student encounters a question they're unsure about (from textbook, mock test, or
previous exam)
3. Student taps "Scan Question" button and takes a clear photo of the question
4. AI processes the image (takes 15-30 seconds):
Recognizes the question text
Analyzes the content
Identifies the exact topic (e.g., "Organic Chemistry - Reaction Mechanisms")
5. App displays topic identification with confidence score (e.g., "This is an SN1
Reaction question - 95% confident")
6. Student is shown 3 key concepts related to this topic (e.g., "What is SN1?", "Rate-
Determining Step", "Carbocation Stability")
7. Student can tap on any concept to see detailed explanation with diagrams,
mechanisms, and worked examples (takes 2-3 minutes to read)
8. App recommends 4-5 similar practice problems related to this topic
9. Student attempts these practice problems within the app
10. AI evaluates each attempt and provides:
Correct/Incorrect feedback
Detailed explanation if wrong
Encouragement if right
Option to see solution if student is stuck
 11. Student completes practice problems and gets a performance score (e.g., "You
scored 4/5 on SN1 questions")
12. App learns this attempt and adds "SN1 Reactions" to student's weakness list
13. Student continues studying other topics and repeatedly scans questions they're
unsure about
14. Over 2-3 weeks, app builds a profile of student's weak topics (e.g., "Weak in: Organic
Reactions, Weak in: Thermodynamics, Strong in: Stoichiometry")
15. App sends personalized recommendation (e.g., "Based on your scans, you're weak
in 3 topics. Spend the next week on Organic Reactions & Thermodynamics")
16. Student follows the personalized revision plan instead of randomly choosing
topics
17. Student's exam performance improves because they focused on actual weak areas
instead of wasting time on strong areas
18. Student feels more confident because app shows clear progress ("You improved
25% on Organic Reactions - great progress!")
19. Student uses app throughout the year during all revision phases (weekly practice,
unit tests, mock exams, final exam prep)
20. By exam day, student is confident because app confirmed they've practiced weak
topics thoroughly and achieved high accuracy

Self-Assessment for Design

Rating: 3 points
Justification: The use of AI is a good fit for the solution. The new user experience is clearly
documented showing how users will be better served than they are today. The 20-step
process contrasts sharply with the old 10-step process, highlighting that students now get:
(a) instant topic identification instead of waiting for teachers, (b) personalized
recommendations instead of generic materials, (c) immediate feedback instead of 5-7 day
delays, and (d) clear weak topic prioritization instead of guesswork.

SECTION 7: DATA
7.1 What Data Will You Need to Train Your AI Solution?
The core of our solution is an image classification model that recognizes the topic of a
chemistry question from its photograph. To train this model, we need:
Training Data Required:
Chemistry question images (photos or scans of actual questions from textbooks,
mock tests, and exams)
Each image must be labeled with its topic (e.g., "Organic Chemistry - SN1
Reactions", "Physical Chemistry - Thermodynamics", "Inorganic Chemistry -
Transition Metals")
Minimum 150-200 images per topic (we're starting with 5 major topics, so 750-1000
images total)
Variety in image quality (clear photos, slightly blurry photos, photos at different
angles) to make the model robust
Secondary Data (for personalization):
Student quiz attempt data (which problems they attempt, correct/incorrect, time
taken)
Topic difficulty ratings (which topics students commonly struggle with)

7.2 Data Sources

 Who Do You
Data Where Will the Ethical
Owns Have
Need Data Come Considerat
the Permission
ed From? ions
Data? ?
1. Photos of
NCERT textbook NCERT Yes - NCERT
Chem questions (public is public
Educationa
istry (Chapters 1-15) 2. domain) domain. Yes
l use is fair.
questi Previous year , State - exam
Will not
on board exam Board papers are
sell/redistri
image papers (available (educati published
bute. Will
s on PuStack, onal for student
credit
(Have Vedantu, use), practice. Yes
sources.
) YouTube) 3. Our Our - school gave
school's mock test school permission
papers
Subjective,
Topic Manually label Us (our but
Yes - we
labels each image based team standardiz
created the
(Have on NCERT chapter labeling ed by
labels
) classifications ) NCERT
structure
Privacy:
Store
Stude Yes - with
Student anonymous
nt parental
Collected from 20- s ly. Consent:
quiz consent.
30 students using (anony Parental
data We'll use
our app during mous approval
(Want anonymous
testing phase particip taken. Data
/Need ID instead of
ation) security:
) real names
Encrypted
storage.
Topic Collected from Student Yes - Non-
diffic student surveys feedbac anonymous identifying.
ulty ("rate this topic's k and Voluntary
rating difficulty 1-5") (anony voluntary participati
s and performance mous) on.
(Want data
 Who Do You
Data Where Will the Ethical
Owns Have
Need Data Come Considerat
the Permission
ed From? ions
Data? ?
/Need
)
Conce
pt
Write our own
expla Yes - our
explanations Paraphrase
natio Our original
based on NCERT + instead of
n team writing or
chemistry plagiarize.
datab created paraphrased
websites (Toppr, Credit
ase content + cited
Khan Academy sources.
(Nice sources
concepts)
to
have)
Practi
ce
Yes - for Fair use for
probl Public
NCERT solved educational educationa
em educati
examples, board purpose, l purpose.
datab onal
exam solutions adapted Non-
ase resourc
(adapted) with our commercia
(Nice es
explanations l.
to
have)

Self-Assessment for Data

Rating: 3 points
Justification: Relevant data to train the AI model have been identified as well as how the
data will be sourced or collected. There is evidence that the dataset will be balanced (we'll
collect equal number of images per topic topic), and that safety and privacy have been
considered (anonymous student IDs, parental consent, encrypted storage, no
redistribution, educational-use-only).
SECTION 8: PROTOTYPE
8.1 Which AI Tool Will You Use to Build Your Prototype?
Selected Tool: ML4All

Why ML4All?
Free and cloud-based - No installation; access from any browser
No coding required - Perfect for high school students without programming
background
Beginner-friendly - Drag-and-drop interface; very intuitive
Built for image classification - Our main need; ML4All specializes in this
Trains models quickly - Can train on 200 images in 5-10 minutes
Good accuracy - Achieves 85-95% accuracy for well-defined categories
Export as API - Can connect to a simple app/website after training
Documentation and tutorials - Many YouTube videos and guides available

Backup tools (if needed):

Google's Teachable Machine (even simpler, but fewer customization options)
TensorFlow Lite (more powerful but requires coding - Plan B only)

8.2 How Will You Build Your Prototype?

4-Week Prototype Development Timeline:
Week 1: Data Preparation

Collect ~800 chemistry question images from NCERT + exam papers

Label each image with its topic (5 categories: Organic, Physical, Inorganic, General,
Analytical)
Organize images in folders by topic
Week 2: ML4All Model Training
Create account on ML4All
Upload labeled images (200 images per topic)
Train image classification model
Test accuracy (goal: >85%)
Make adjustments if needed (more images, adjust parameters)

Week 3: Build User Interface

Create simple HTML/CSS frontend where users can:
Upload a photo
See the predicted topic
See related practice problems
Track their weak topics
Option: Use Google Forms or Jotform if coding is too complex
Integrate ML4All API with the interface
Week 4: Connect Everything & Test
Connect frontend to ML4All model via API
Test end-to-end: Upload photo → Get prediction → Show practice problems
Collect feedback from 5-10 test users
Fix bugs and improve user experience

8.3 What Decisions Will Your Tool Generate and What

Action Happens Next?
Decisions Made by AI:
1. Topic Identification Decision
AI receives: Chemistry question image
AI decides: "This is a question on [Topic Name]"
Confidence score: 94%
Action: Display the topic + show 3-5 key concepts for this topic
2. Weak Topic Detection Decision
AI receives: Student's photo history (20+ scanned questions over 2 weeks)
AI analyzes: "Student frequently scans questions from Organic Chemistry"
AI decides: "Organic Chemistry is a weak topic for this student"
Action: Send notification - "We noticed you struggle with Organic Chemistry.
Let's focus on this next week."
3. Practice Problem Recommendation Decision
AI receives: Topic identification result
AI decides: Which 4-5 practice problems best match this topic
Action: Display recommended problems in order of increasing difficulty
4. Progress Evaluation Decision
AI receives: Student's practice attempt data (5 attempts on SN1 reactions)
AI decides: "Student achieved 4/5 correct (80% accuracy)"
AI decides: "Student is ready to move to next topic" OR "Student needs more
practice on this topic"
Action: Send encouragement message + next recommendation
5. Personalized Study Plan Decision
AI receives: Week's worth of data (which topics scanned, performance on each
topic)
AI decides: Priority ranking of topics to focus on next week
Action: Send study plan - "Next week: Focus on Thermodynamics (weak) and
Kinetics (very weak). Less time needed on Stoichiometry (strong)."

Self-Assessment for Prototype

Rating: 3 points
Justification: A prototype for the solution has been created and successfully trained to meet
users' requirements. We achieved 91% accuracy on topic classification over 800 labeled
images across 5 topics, which exceeds our goal of 85%. The prototype successfully performs
its core function: taking a photo of a chemistry question and correctly identifying the topic
with high confidence. Users in testing confirmed that the topic identification is accurate
and the recommended practice problems are relevant.
SECTION 9: TESTING
9.1 Who Are the Users Who Tested the Prototype?
We conducted testing with 8 high school students:
1. Student Tester A (10th Grade, Chemistry, Female) - Currently using coaching
2. Student Tester B (10th Grade, Chemistry, Male) - No coaching (self-study)
3. Student Tester C (11th Grade, Mixed Subjects, Female) - Preparing for NEET
4. Student Tester D (11th Grade, Mathematics focus, Male) - Will use for math photos
too
5. Student Tester E (12th Grade, preparing for JEE, Female) - Advanced student
6. Student Tester F (12th Grade, Chemistry, Male) - Average student, struggles with
concepts
7. Student Tester G (10th Grade, Science, Female) - Less tech-savvy
8. Student Tester H (11th Grade, Chemistry, Male) - Already using other AI tools

Diversity: Mix of genders, grades, coaching status, tech comfort, and prior experience with
AI tools

9.2 Observations of Users as They Tested the Solution

Session 1 - Testing Day 1 (2 hours, 4 students)
Observation #1: All 4 students understood the app within 30 seconds. They immediately
took photos of random chemistry questions from a worksheet without needing
instructions.

Observation #2: When AI correctly identified a topic (e.g., "SN1 Reactions"), students
reacted with surprise and delight ("Wow, it got it right!"). This suggests the feature is novel
and impressive to them.
Observation #3: Student A took a photo of a physics question by mistake. AI correctly said
"This is not a chemistry question." Students found this verification helpful - "It prevents me
from getting confused between subjects."
Observation #4: Student B tried testing the tool with a badly written/blurry photo. AI
showed lower confidence (71%) but still predicted the topic correctly. Student reacted: "It's
honest about uncertainty - I like that."

Observation #5: All students wanted to see the practice problems immediately after topic
identification. Three said, "Okay, now show me similar problems to practice." This confirms
our design choice to display recommendations right after topic ID.
Observation #6: Student G (less tech-savvy) needed a 20-second clarification on "where to
tap to upload photo" but then used the tool independently. The interface is intuitive enough
for non-tech users.
Observation #7: When practice problems appeared, students engaged with them. 7 out of 8
spent 5-10 minutes attempting the problems without prompting.
Observation #8: Student C noticed the explanations for wrong answers and said, "This is
better than just seeing 'incorrect' on my mock test." She spent extra time reading
explanations even when she got answers right.

Session 2 - Testing Day 2 (2.5 hours, remaining 4 students)

Observation #9: The second batch of testers had similar positive reactions. One mentioned,
"I wish my teacher was this fast at giving feedback."
Observation #10: Student D (math focus) asked, "Can I upload math problems too?" We
clarified it's chemistry-only right now, but he expressed interest in a multi-subject version.

Observation #11: Students were curious about how AI learned: "How does it know this is an
SN1 question?" This shows they're thinking about AI's mechanism and trust it more when
they understand it.
Observation #12: No student complained about accuracy. All photos correctly identified
topics, which is impressive since our training data was diverse (different photo qualities,
handwriting styles, textbook vs exam papers).
Observation #13: One student (F) who struggles with concepts said the explanations
"finally make sense. My textbook explanation is too technical." This is exactly the user
we're trying to help.

9.3 User Feedback Grid

 What Needs
What Works Questions? Ideas
to Change
Topic Will this
identification is work with
The app is
highly accurate - any Let us use this
chemistry-
Correctly chemistry for all subjects
only; add math
identified 31/32 textbook, or in the future
and physics
photos (97% just
accuracy) NCERT?
Explanations
Instant feedback
are sometimes Make a "quick
- Predictions in
too long (5-10 How does summary"
<20 seconds;
minutes to the AI know mode with 2-
much faster than
read all); the topic? minute
waiting for
could be explanations
teacher
shorter
Practice
problems are Why does it
Sometimes we
relevant - sometimes
want to Add a search
Students said show low
SEARCH for a bar: "Show me
90% of confidence
specific topic SN1 Reaction
recommended (71%
instead of problems"
problems instead of
taking a photo
matched the 95%)?
topic well
Concept Need more
Expand our
explanations practice
database - need
are clear - problems per Can it
300+ problems
Multiple students topic explain the
per topic
said explanations (sometimes photo too?
instead of
are "better than only 2-3
current 50
textbook" available)
Encouragement The progress How is my Simplify
messages are tracking data being dashboard -
motivating - interface is stored? Is it show only 3 key
Students liked hard to private? metrics instead
"You improved understand of 10
 What Needs
What Works Questions? Ideas
to Change
20% on this (too many
topic" messages graphs)
Learned quickly Some topics Add simple
that bad photos need better teacher
= low confidence visual Can dashboard:
- Students explanations teachers allows teachers
appreciated (diagrams/ani see our to see which
honest mations progress? students
confidence instead of text struggle with
scores only) which topics
Add a social
Helped them Would be nice feature: "You're
prioritize which to see Will this in 75th
topics to focus comparison help me percentile on
on - "Finally with pass my this topic"
know where to classmates exam? (motivating but
focus" (anonymously) non-
competitive)
Allow offline
access - many
students study
without
internet

9.4 Urgent Improvements Needed Now (Before Final

Submission)
Based on user feedback, we need to fix these BEFORE showing to teachers/using in
production:
1. Expand practice problem database (currently 50 problems per topic → need 150+)
Estimated time: 3 hours
Responsible: Data Manager
Impact: High (critical for users' core need)
2. Simplify the progress dashboard (too many graphs and metrics confusing users)
Estimated time: 2 hours
Responsible: UI Design Lead
Impact: High (users need to understand progress at a glance)
 3. Add "Quick Summary" mode for explanations (5-10 minute explanations too long
during study)
Estimated time: 4 hours
Responsible: AI Developer
Impact: Medium (nice-to-have but users specifically requested)
4. Fix confidence score display (show why confidence is low, e.g., "Multiple possible
topics: SN1 or SN2")
Estimated time: 1 hour
Responsible: AI Developer
Impact: Medium (users want to understand uncertainty)
5. Test with 10 more students to confirm improvements work
Estimated time: 4 hours (2 hours testing + 2 hours feedback analysis)
Responsible: Testing Lead
Impact: Critical (ensure solution works for diverse users)

9.5 Improvements for Later (Nice-to-Have Features)

These features are useful but not critical; we can add them in future versions:
1. Multi-subject support (math, physics, biology in addition to chemistry)
2. Teacher dashboard - teachers can see which students struggle with which topics
3. Social/Anonymous Leaderboard - "You're in 75th percentile on Organic Chemistry"
4. Video explanations - animated videos instead of text-only explanations
5. Offline mode - use app without internet (download content first)
6. Voice input - "Hey AI, explain SN1 reactions" instead of typing
7. Integration with exam dates - "Your exam is 4 weeks away; spend more time on
weak topics"
8. Parent notifications - parents can see their child's progress without accessing the
app directly
9. Gamification features - badges, streaks, daily challenges for motivation
10. Concept roadmap visualization - "You've mastered 40% of Organic Chemistry;
here's what's left"

Self-Assessment for Testing

Rating: 3 points

Justification: A prototype has been tested with a fair representation of users (8 testers,
diverse by grade/coaching status/tech comfort) and all tasks in this section have been
completed. We conducted 2 testing sessions, collected detailed observations of user
behavior, gathered structured feedback, identified 5 urgent improvements, and listed 10
nice-to-have features for future versions. User feedback confirms the core hypothesis:
students find the tool helpful, accurate, and prefer it over waiting for teacher feedback.
SECTION 10: TEAM COLLABORATION
10.1 How Did You Actively Work with Others and With
Stakeholders?
Team Collaboration Strategies
1. Clear Role Definition & Accountability
Each team member had a specific role (Project Lead, Data Manager, AI Developer, UI
Designer, Testing Lead, Documentation Lead)
Weekly meetings to ensure everyone stayed on track
Project Lead monitored progress and followed up if someone fell behind
Everyone felt ownership over their part

Example: When Data Manager fell behind on collecting images (had 400 out of 800), we
helped them by each contributing 50 images. Everyone pitched in because we understood
the critical importance.
2. Daily Communication via WhatsApp
Quick status updates every morning ("Data collection 60% done", "ML4All training
successful")
Questions answered within hours instead of waiting for weekly meetings
Celebration of wins ("Model achieved 91% accuracy - go team!")
This kept everyone motivated and aligned

3. Collaborative Decision-Making
When selecting which topic to focus on (revision problem), we held a 1-hour
brainstorming session where everyone could voice opinions
Used priority grid to make selection objective; not just one person's choice
Resulted in team buy-in because everyone felt heard
4. Peer Learning & Knowledge Sharing

AI Developer taught everyone how to use ML4All (30-min tutorial)

Now everyone could check model accuracy, not just the AI Developer
If AI Developer was absent, others could continue work
Reduced single-person dependency
5. Testing Phase Collaboration
Testing Lead conducted sessions, but entire team observed
All team members interviewed test users (not just Testing Lead)
Everyone helped analyze feedback, not just Testing Lead
Resulted in richer insights and team understanding of what users really needed

6. Conflict Resolution
When we wanted to include a multi-subject feature but project timeline didn't allow:
First, everyone stated their viewpoint in a calm meeting
 We prioritized based on user feedback (chemistry-only was sufficient)
Compromised by noting multi-subject as a "future version" feature
Everyone accepted the decision because reasoning was transparent
7. Stakeholder Engagement

Students (our users): Conducted interviews, invited 8 to testing, incorporated their

feedback into next iteration
Teacher/Mentor: Met with teacher bi-weekly to show progress, got approval for
using school's mock test papers, received guidance on technical approach
School administration: Got permission to test with students, mentioned our project
in school newsletter
8. Supporting Each Other During Challenges
When Data Manager was sick for 2 days, UI Designer and Project Lead covered
image labeling
When AI Developer struggled with ML4All API integration, we brought in a senior
student who had used it before
Nobody said "that's not my job" - team mindset

9. Celebrating Milestones
Each phase completion (Problem Definition done, Model trained, Testing complete)
was celebrated
Team photos after testing sessions
Group message: "We did it! 8 students tested and loved our solution!"
Motivation came from celebrating together, not individual recognition
10. Documentation & Knowledge Transfer

Documentation Lead maintained a shared Google Doc with all decisions, meeting
notes, and learnings
If someone graduated or left the team, new member could catch up quickly
At project end, comprehensive knowledge base for future teams

Self-Assessment for Team Collaboration

Rating: 3 points
Justification: Effective team collaboration and communication among peers and
stakeholders is clearly documented. Team members had distinct roles, communicated
daily, made collaborative decisions, supported each other during challenges, and actively
engaged users and teachers as stakeholders. The team operated as a cohesive unit from
problem definition through testing, with clear documentation of how collaboration
contributed to project success.
SECTION 11: INDIVIDUAL LEARNING
REFLECTION
11.1 Team Reflections
Team Member 1: ________________________
What surprised me during this project:
The biggest surprise was how much we didn't understand about our users before
conducting interviews. We initially thought the solution should include features like
animated videos and gamification, but when we talked to real students, they said they just
wanted instant feedback and clear guidance on which topics to focus on. This taught me
that real user research is 100x more valuable than our guesses.

Another surprise was how quickly AI could work. I thought training an image
classification model would take weeks and require advanced coding. Instead, ML4All
trained our model to 91% accuracy in just 4 days and required zero coding.
The hardest part was the data preparation phase. Labeling 800 images manually was
tedious, but I realized this "boring" work is what makes the model accurate. Quality data >
fancy code.
Learning: Before building anything, talk to your actual users. Your assumptions will be
wrong.

Team Member 2: ________________________

What surprised me during this project:

I was shocked that students were so eager to help us test our solution. I thought we'd
have to beg for testers, but 8 students volunteered immediately. This showed me that if
you're solving a real problem that affects people, they'll be invested in your success.
Another learning: Teaching is not a given. When I taught the team about ML4All, I had to
explain it 3 different ways before everyone understood. This made me empathize with
teachers who struggle when a single explanation doesn't work for everyone. (It also made
me appreciate why our AI explaining wrong answers in multiple ways is so valuable!)
The frustrating part was when the model made confident wrong predictions on a few
images. I wanted to give up and blame the data, but our team investigated and found I had
mislabeled a few images. Responsibility over blame - we owned our mistakes.

Learning: In AI projects, focus on data quality, not just model complexity. A simple model
with great data beats a complex model with messy data.
Team Member 3: ________________________
What surprised me during this project:
I was surprised how collaborative the AI community is. When our team got stuck on API
integration, I posted on a forum and got 5 detailed responses within 24 hours. People
genuinely wanted to help.
Another surprise: The project is not "done" after testing. I thought once we got 91%
accuracy, we were finished. But testing revealed so many improvements needed: expand
database, simplify dashboard, add quick summaries, etc. I realized that real software is
constantly evolving based on user feedback.

What shocked me most was the responsibility of building a tool that students might trust.
When a student said "finally I understand this concept," I felt the weight of potentially
influencing their exam performance and confidence. This is not just a school project - it
could genuinely help someone.
Learning: AI tools are only as good as their real-world impact. Accuracy numbers don't
matter if users don't actually use your solution or if it doesn't improve their lives.

Team Member 4: ________________________

What surprised me during this project:
The biggest lesson was about scope management. We initially wanted to include features
like multi-subject support, teacher dashboard, and social leaderboard. But with limited time
and resources, we had to say "no" to good ideas and focus on the core problem. This was
hard because we wanted to include everything, but I learned that done > perfect.

Another surprise: User feedback is brutally honest and incredibly valuable. When
testers said "your explanations are too long," I initially felt defensive (we worked hard on
those explanations!). But then I realized they were right - in real-world usage, students want
quick summaries, not 10-minute reads. Feedback isn't criticism; it's a gift.
I also learned that technical skills matter less than problem-solving skills. I don't know
advanced coding, but I could identify bottlenecks, help teammates, coordinate logistics,
and communicate with stakeholders. Leadership doesn't require being the smartest person.
Learning: Ship something users can actually use. Perfect is the enemy of useful.

Team Member 5: ________________________

What surprised me during this project:
I learned that problems that seem obvious aren't always solved. Revision effectiveness is
a well-known problem in education, yet our solution (instant AI-based feedback) doesn't
exist widely. This taught me that good ideas are worth pursuing even if similar problems are
worldwide.

Another reflection: Teamwork multiplies effort. If I had done this project alone, I would
have given up during the tedious image-labeling phase. But because I had teammates, I
didn't want to let them down. We pushed each other to finish tasks and celebrate milestones
together.
The most humbling moment was during testing when a struggling student used our tool and
suddenly understood a concept they'd been stuck on for weeks. That moment made all the
boring meetings and data-labeling worthwhile.

Learning: Solve real problems for real people. That's the best motivation.

Team Member 6: ________________________

What surprised me during this project:

I realized that documentation is not boring busywork. Keeping detailed meeting notes
meant we could recall why we made certain decisions later. Without documentation, we
would have re-debated settled issues.
Another learning: Failure is part of the process. Our initial model had only 73% accuracy
because we mislabeled images and had imbalanced data. Instead of giving up, we fixed the
data and accuracy jumped to 91%. I learned that technical failures usually indicate data
problems, not model problems.
I was surprised by how important the user interface is. We could have built a powerful
backend, but if the interface is confusing, users won't use it. This project taught me that User
Experience (UX) is as important as technical performance.

Learning: In any AI project, invest time in data quality, documentation, and user interface.
The "boring" parts actually determine success.

Self-Assessment for Individual Learning Reflection

Rating: 3 points
Justification: Each team member presents a reflective and insightful account of their
learning during the project. Reflections go beyond surface-level observations and include
deep lessons about: user research importance, data quality > model complexity, scope
management, feedback reception, teamwork value, and the importance of solving real
problems for real people. Each reflection identifies a specific surprise/learning moment and
connects it to a broader lesson applicable beyond this project.

SECTION 12: VIDEO LINK

Enter the URL of your team video:

[Your YouTube link will go here - e.g., [Link]

Enter the password (if any):
[If the video is unlisted or private, add password here]
APPENDIX: ASSESSMENT RUBRIC
 Our Po
Criteria Rati int Notes
ng s
Problem
Detailed research, specific problem
Definitio 3/3 ✓
identified, user interviews conducted
n
Comprehensive user research with 3
The
3/3 ✓ interviews, empathy map, detailed
Users
workflow analysis
5 creative ideas, systematic priority grid
Brainsto
3/3 ✓ evaluation, compelling solution selected
rming
with strong justification
AI is perfect fit for solution, new user
experience clearly documented with 20-
Design 3/3 ✓
step process showing dramatic
improvement over old process
Relevant data identified (800 chemistry
question images), sourcing strategy clear,
Data 3/3 ✓
balanced dataset confirmed, privacy/safety
considered
Prototyp Prototype created and successfully trained
3/3 ✓
e (91% accuracy), meets user requirements
Tested with 8 diverse users, clear
Testing 3/3 ✓ improvements made, urgent fixes
identified, future features documented
Effective collaboration documented with
Team
clear role assignments, peer learning,
Collabor 3/3 ✓
conflict resolution, stakeholder
ation
engagement
Individu
Each team member's reflective account
al
3/3 ✓ shows genuine learning and insights
Learnin
beyond surface-level observations
g
Video 3/3 ✓ 3-minute video with clear communication,
Presenta (exp demonstrations, passion, good
 Our Po
Criteria Rati int Notes
ng s
tion ecte sound/image quality
d)

TOTAL
LOGBOO 27/ Complete and comprehensive
K 27 documentation
POINTS

Project Status: Ready for Final Submission ✓

All sections completed with thorough detail. Prototype tested and refined. Team
collaboration excellent. Ready to submit to IBM EdTech platform.