Problem Statement:

A Digital Learning Platform Empowering Individuals with Disabilities to Access Inclusive

Education

Description:
Many individuals with physical, sensory, or cognitive disabilities face significant barriers to
accessing quality education and skill development opportunities. Traditional learning
environments often lack accessibility features, adaptive tools, and personalized learning
paths that cater to their unique needs.
This project aims to design and develop an inclusive e-learning platform that enables people
with disabilities to learn effectively and independently. The platform will integrate
accessibility features such as screen reader compatibility, speech-to-text and text-to-speech
tools, sign language video support, adjustable font sizes, high-contrast modes, and easy
navigation. It will also include interactive lessons, gamified learning experiences, and
community support to foster engagement and confidence.
By leveraging technology and universal design principles, this platform will bridge the
educational gap, promote digital inclusion, and empower especially abled individuals to
enhance their knowledge, skills, and career opportunities.

Domain:
Smart Education
Problem Statement:
Adaptive Learning Platform that Personalizes Curriculum Paths Based on Knowledge Gaps,
Learning Styles, Pace, and Real-Time Emotional Engagement

Description
As education increasingly moves toward digital and personalized learning, students face
challenges in receiving content that suits their unique learning needs. Current learning
platforms tend to follow a one-size-fits-all approach, relying on static assessments that fail to
adapt to individual progress and learning styles. This limits their ability to fully engage
students and address specific knowledge gaps, which can affect their overall learning
outcomes.

The goal of this project is to develop an adaptive learning platform that assesses each
student’s unique learning profile, including their knowledge gaps, preferred learning style,
and learning speed. Based on this information, the platform will create customized
curriculum paths with interactive lessons, videos, and practice problems that adjust in real
time according to the student’s progress. The system should be able to continuously monitor
and adapt to the student’s changing needs, providing a truly personalized learning journey.

This innovative learning platform customizes the educational experience by analysing each
student’s knowledge gaps, preferred learning style, and pace. It dynamically generates
individualized curriculum paths, ensuring students receive content tailored to their specific
needs.

The system goes beyond static learning paths by offering interactive content, such as videos,
quizzes, and practice problems, which adjust in real-time based on a student’s performance.
As the student progresses, the platform automatically adapts the difficulty level of tasks,
keeping them challenged but not overwhelmed.

A unique aspect of this platform is the integration of emotion-sensing technology, which

monitors real-time emotional engagement and frustration levels using tools like facial
recognition or wearable sensors. By assessing how students feel during their learning process,
the platform fine-tunes the content to maintain motivation and avoid frustration, ensuring a
balanced and effective learning experience.

The platform should have the potential to provide a truly personalized learning
experience that helps students progress at their own pace, stay motivated, and tackle
knowledge gaps effectively.
Domain
Smart Education
Problem Statement:
Develop an AI Assistant that can effectively answer questions based on provided
documentation.

Description:
The objective is to develop an intelligent virtual assistant capable of parsing, comprehending,
and referencing large volumes of textual documentation to answer user queries. This system
should support natural language processing (NLP) to interpret user questions and identify the
most relevant sections of the documentation that address the query.
The assistant must:
Extract Information: Automatically scan and analyze provided documents, understanding key
terms, definitions, procedures, and concepts.
Understand Queries: Process user questions written in natural language, even if phrased
ambiguously or imprecisely, and interpret the intent behind them.
Contextual Relevance: Ensure that the assistant responds with information relevant to the
user’s specific query, while considering the broader context and any potential nuances in the
documentation.
Provide Clear Responses: Generate concise, accurate, and easy-to-understand answers that
are directly related to the question, along with references to relevant sections or excerpts
from the documentation when necessary.
Learning and Adaptation: The system should also adapt over time to improve its ability to
answer complex questions by learning from previous interactions, user feedback, or by
continuously updating itself with new or modified documentation.
The AI assistant would be particularly useful in environments such as customer support,
software documentation, technical manuals, or any knowledge-based system where users
frequently need assistance navigating large amounts of text-based information. By
automating the process of querying documentation, the assistant enhances efficiency,
accuracy, and user experience, allowing users to quickly find the information they need
without manually searching through documents.

Domain:
Large Language Models (LLMs) and Generative AI
Problem Statement:
AI-Driven Infrastructure Diagram & IaC Code Generator

Description
Cloud computing has become integral to modern software development and deployment.
Infrastructure as Code (IaC) tools like Terraform, AWS CloudFormation, and Azure ARM
templates are widely used to automate the provisioning of cloud resources. However,
creating and managing cloud infrastructure, especially for complex systems, is a time-
consuming and error-prone process. In addition, visually representing this infrastructure in
the form of diagrams that can be easily understood by teams and stakeholders is a crucial
yet often tedious task.
Challenge:
Your challenge is to build an AI-based solution that can:

Automatically generate cloud infrastructure diagrams based on textual descriptions or

predefined templates. The system should be able to identify key components such as Virtual
Private Clouds (VPCs), EC2 instances, RDS databases, Load Balancers, and their relationships,
and represent them visually in a clear and understandable format.

Generate corresponding Infrastructure as Code (IaC) for the described infrastructure. The
system should support multiple IaC frameworks like Terraform, AWS CloudFormation,
and Pulumi, and ensure the generated code is syntactically correct, optimized, and ready for
deployment. Additionally, the solution should maintain high security standards and ensure
that the generated IaC code adheres to industry compliance regulations.

Allow users to manually edit the generated diagrams. While automatic generation is
valuable, sometimes an infrastructure design needs to be adjusted to meet specific business
requirements or team preferences. The system should offer functionality to modify the
diagram directly by adding, removing, or reconfiguring resources. It is critical that these
manual changes are reflected in the generated IaC code in real-time.

Ensure seamless synchronization between the diagram and IaC code. If the user edits the
diagram manually, the system should automatically update the corresponding IaC code to
reflect the changes. Similarly, if modifications are made to the IaC code, the system should
ensure the diagram is updated to reflect those changes. This bi-directional synchronization
should maintain consistency and prevent discrepancies.

The system should also incorporate security standards and compliance checks to ensure
that the generated infrastructure follows best practices for security, such as enforcing the
principle of least privilege, ensuring encryption at rest and in transit, and adhering to
relevant industry standards like ISO 27001, GDPR, and HIPAA.
Key Features:
The solution should be capable of text-to-diagram generation, allowing users to input a
description of the desired infrastructure (e.g., "Create a VPC with two subnets and an EC2
instance running in a public subnet") and receive an automatically generated diagram that
represents the infrastructure. Along with the diagram, the system should produce
corresponding IaC code (e.g., Terraform, CloudFormation) that describes the same
resources and their configurations.

In addition, users must be able to manually edit the diagram, including adding, removing, or
adjusting components. These changes should be reflected in the IaC code in real-time,
ensuring the accuracy and consistency between both the diagram and the infrastructure
code. The solution should be capable of offering best practice recommendations and
ensuring compliance with security guidelines, such as setting up appropriate IAM roles,
configuring secure subnets, and applying encryption where needed.

To enhance the user experience, the interface should be intuitive and user-friendly. It
should support drag-and-drop functionality for adding cloud resources and allow for live
editing and updates. Furthermore, the system should provide real-time feedback on any
potential issues or best practices for configuring resources, ensuring that users are aware of
security implications or compliance gaps.

By the end of the hackathon, the solution should empower cloud architects, DevOps
engineers, and developers to quickly design, deploy, and manage cloud infrastructure. The
tool should automate the tedious tasks of diagram creation and IaC code writing, saving
time and reducing the potential for human error. At the same time, it should enable teams
to maintain flexibility by allowing manual edits to the generated diagrams while keeping the
IaC code consistent. Importantly, the solution should ensure that the generated
infrastructure adheres to industry security standards and compliance regulations, enabling
organizations to safely and efficiently manage their cloud environments.

Domain
Large Language Models (LLMs) and Generative AI
Problem Statement:
AI-Driven Demand Forecasting: Leveraging Generative AI for Unified Retail Data Insights

Description
Retailers are often overwhelmed by the massive amounts of data they collect across various
formats: sales data in CSV/TSV files, inventory logs in Excel, marketing data in JSON, weather
reports, foot traffic logs, and even images of products shared on social media. These data
types are usually unstructured, inconsistent, and stored without organization, making it
difficult to extract meaningful insights — especially for predicting demand for seasonal
products.
In this hackathon, your challenge is to build a Generative AI (GenAI)-powered system that
can automatically process, clean, and integrate diverse data formats to deliver actionable
insights for retail demand forecasting.
Here’s what your system should do:
• Data Understanding and Tagging:
o Use Generative AI techniques to process incoming files (CSV, JSON, Excel, image
files, logs) and automatically tag and categorize the data based on a retail-specific
ontology.
▪ For example: tagging sales data with product categories, inventory with
stock levels, weather data with regions experiencing snowfall, and images
with relevant seasonal trends.

• Use-Case-Driven Data Selection:

o When a user provides a specific query, like “Predict demand for winter jackets in
snowy regions,” your AI system should intelligently filter and select only the data
that’s relevant to the query.
o For instance, it should extract weather data for regions with snow, historical sales
data for winter jackets, and marketing campaign data targeting winter
promotions.

• Data Preprocessing and Cleaning:

o Apply Generative AI to automatically clean and preprocess raw data in different
formats:
▪ CSV/TSV/Excel: Handle missing values, standardize dates, normalize units,
and align region/product names.
▪ JSON: Flatten nested structures and convert campaign details into usable
attributes.
▪ Logs: Extract meaningful metrics like visitor counts and time-based
patterns.
▪ Images: Use image recognition models to extract tags (e.g., jackets, snow,
boots) and associate them with seasonal trends.
• Ontology-Aware Data Integration:
 o Use an ontology-driven approach to intelligently merge diverse datasets:
▪ Combine sales data with weather insights (e.g., how snow affects winter
product purchases).
▪ Merge marketing campaign data with foot traffic logs to understand the
impact of promotions on store visits.
▪ Integrate images with sales data to identify customer interest in certain
product categories (e.g., winter jackets in snowy settings).
• Output Actionable Insights:
o Generate a clean, unified dataset ready for analysis, providing insights on how
different factors (weather, marketing campaigns, store foot traffic, and product
interest) impact seasonal product demand.
o Build intuitive dashboards or visualizations that show trends, correlations, and
actionable insights, such as predicting future demand for specific products in
snowy regions.

This challenge involves building a Generative AI system that processes and integrates diverse
data formats (CSV, TSV, JSON, Excel, images, logs). The system should include a preprocessing
pipeline that handles missing values, normalizes units, flattens structures, and automatically
tags data. It will create a unified, ontology-aware dataset combining sales, weather,
marketing, foot traffic, and images for predictive demand forecasting. Participants will
generate insights or visualizations demonstrating how this data integration enhances
forecasting accuracy, optimizing inventory and marketing strategies. The focus is on using
Generative AI to automate data understanding, tagging, and forecasting, enabling retailers to
make smarter, data-driven decisions.

Domain
Large Language Models (LLMs) and Generative AI
Problem Statement:
Develop an application aimed at supporting rescue teams and individuals in disaster-affected
regions by delivering real-time situational awareness, resource mapping, and advanced
capabilities to locate people trapped under debris or in areas otherwise inaccessible to visual
detection.

Description:
In the aftermath of a disaster, rescue operations often face the critical challenge of identifying
and reaching survivors trapped in areas that are difficult to access or not visible to rescuers.
This application is designed to aid both rescue teams and individuals in disaster-affected
zones by providing a suite of real-time tools and capabilities. Key features include:
Real-Time Information: Deliver up-to-the-minute data regarding disaster conditions,
including affected areas, available resources, and survivor status.
Resource Mapping: Enable efficient tracking and management of resources like food, medical
supplies, and rescue equipment, ensuring that teams can prioritize and distribute aid
effectively.
Survivor Detection: Using advanced technologies (such as thermal imaging, drones, or
sensors), the app can help locate survivors trapped under rubble or in hidden locations that
are not visible through traditional search methods.
Coordination & Communication: Facilitate seamless communication between rescue teams,
local authorities, and survivors to ensure efficient coordination of efforts.
AI-Based Damage Assessment: Using artificial intelligence, the app could analyze satellite
imagery, drone footage, or photos uploaded by users to automatically assess the extent of
structural damage. This would help prioritize search and rescue efforts in the most heavily
impacted areas and estimate the medicine supply.
Multi-Language Support: The app could support multiple languages, making it accessible for
all the rescue teams and individuals in different disaster zones, overcoming language barriers
in critical moments.
Survivor Health and Vital Signs Monitoring: Through integration with wearable health
devices (or manual reporting), the app could track the health status of survivors, such as heart
rate, breathing, and other vital signs, to give rescuers critical information about their
condition before reaching them.
Integration with Emergency Services and NGOs: The application could connect seamlessly
with local emergency services, government agencies, and NGOs to enable cross-organization
communication and resource sharing. This would help streamline the distribution of aid,
medical supplies, and personnel to high-priority areas.
The application aims to improve rescue efficiency, reduce response time, and enhance the
likelihood of saving lives by providing critical information and tools in high-pressure, disaster-
stricken environments.

Domain:
Disaster Management
Problem Statement:
Build a platform for evacuation guidance for fire emergencies in residential/commercial
buildings.

Description:
Fire safety norms and protocol are not followed well enough in the country. Building plans
often pose a risk to the people within premises. Unclear guidance and lack of training can lead
to confusion, congestion and unsafe evacuation.
The problem statement demands participants create a platform to utilize smoke detectors,
camera feed and other sensor data with static information (such as floor plans, maps, resident
data) to guide people to the nearest safety point or exit. The guidance can be metted via
various output devices like PA systems, lighting system, digital screens, mobile devices. The
application can also extend to recognize signals that could worsen the disaster.

Domain
Disaster Management
Problem Statement:
Smart Flood Forecasting, Mitigation, and Simulation System

Description
Flooding is one of the most common and devastating natural disasters, causing significant loss
of life, property damage, and disruption to communities worldwide. Traditional flood
forecasting systems often lack the real-time capabilities, accuracy, and dynamic adaptability
required to effectively manage and mitigate flood risks, especially with the increasing
frequency of extreme weather events due to climate change.

The problem is to develop an AI-powered flood forecasting system that integrates multiple
data sources—such as meteorological forecasts, river flow measurements, satellite imagery,
and real-time IoT sensors—to predict flood events with high accuracy. The system should
offer real-time flood risk predictions and early warning alerts to help local authorities and
communities prepare in advance. It must also simulate various flood scenarios, taking into
account different rainfall patterns, storm surges, and river overflow situations, to allow
authorities to visualize the potential impacts of floods and plan effective mitigation measures.

Additionally, the system should recommend flood mitigation strategies based on flood
simulations, such as improving drainage systems, building flood barriers, or optimizing land
use to reduce flood risks. The platform must include a flood simulation tool that allows
emergency responders, urban planners, and local authorities to test different flood
interventions and evaluate their effectiveness in reducing damage before actual flood events
occur. The ability to assess flood impacts on infrastructure, agriculture, and human life is
crucial for efficient resource allocation and disaster management.

As the system will be operating in an environment characterized by high uncertainty and

constantly changing conditions, the challenge also includes ensuring real-time processing of
vast data inputs, addressing data quality and integration issues, and providing user-friendly
interfaces for non-expert stakeholders. Additionally, the system must adapt to changing
climate conditions, continuously improving its predictions and simulations as new data
becomes available.

Domain
Disaster Management
Problem Statement:
Develop an AI powered application designed to assist surgeons by providing precise contour
mapping of surgical areas, identifying sensitive regions, and offering real-time guidance
throughout the procedure.

Description:
This application aims to enhance surgical precision and patient safety by utilizing cutting-edge
technologies such as augmented reality (AR), machine learning (ML), and real-time imaging
analysis. It will offer detailed, interactive contour maps that highlight critical structures (e.g.,
nerves, blood vessels, organs) within the surgical site. Surgeons will be able to visualize these
structures in 3D, helping to avoid damage during operations.
The system will continuously update during the surgery, providing real-time feedback on the
surgeon's actions, and suggesting adjustments when approaching sensitive areas. It will
integrate with existing medical imaging systems (e.g., MRI, CT scans) and use advanced
algorithms to track changes in the surgical environment, ensuring optimal guidance for each
step of the procedure. By incorporating this application, surgical teams can reduce the risk of
human error, improve overall surgical outcomes, and enhance patient safety.

Domain:
Healthcare
Problem Statement:
Develop a system for Predictive Care for Rural/Under-Served Areas

Description:
Develop an AI/ML driven system for healthcare workers operating in rural or remote areas
(e.g., travelling doctors) to:
• Capture patient data
• Analyze and determine the seriousness of disease
• Enable AI/ML-driven predictions
• Allow offline/limited-connectivity operation and sync when connected
In many under-served settings, health workers depend on paper records, manual tracking,
fragmented follow-ups, and lack of analytics support. Patients in rural or underserved areas
often have chronic illnesses (e.g., diabetes, hypertension, heart disease) but lack continuous
monitoring, follow-up and decision-support. Design a solution for a travelling or remote care
provider to capture patient data, track condition progression, provide alerts, and deliver care
recommendations.
• The system should allow input of patient data over time (vital signs, lab results,
symptoms, medications).
• Should work offline (or with intermittent connectivity) given rural constraints, and
sync when connectivity is available.
• Provide predictive analytics (e.g., “based on last 6 months you are at increased risk of
complication X”) or rule-based alerts (e.g., “patient has missed medications for 5 days,
escalate”).
The developed solution could be either single agent or multiple agents.
Example use case:
• Triage assistance and decision support for CT scans/X-rays
• The participants are open to implement any other use case

Domain
Healthcare
Problem Statement:
Early Disease Detection from Medical Imaging

Description
In many resource-limited healthcare settings, access to skilled radiologists and advanced
diagnostic tools is restricted, making it difficult to quickly and accurately detect conditions
like tuberculosis, pneumonia, and bone fractures. Early detection of these diseases is critical,
but manual analysis of medical images is often slow and requires specialized expertise.
Furthermore, healthcare workers may lack the training to interpret complex imaging results
or to discern subtle patterns that suggest early-stage disease.

To address this problem, the goal is to create an AI system that can process medical images—
primarily X-rays and CT scans—and automatically detect early signs of disease. The system
should incorporate advanced techniques like contour mapping to delineate abnormal regions
of the scan, helping to highlight areas of interest and improve the precision of diagnosis.
Additionally, the model should provide explainable reasoning, offering insights into why it
made a particular diagnostic decision (e.g., highlighting specific regions in the image) to
improve user trust and decision-making.

Key elements of the problem include:

1. Accuracy and Sensitivity: The AI model must be highly accurate in identifying early
signs of diseases, reducing both false positives (misidentifying healthy tissue) and false
negatives (missing early-stage abnormalities). Ensuring high sensitivity is critical for
early intervention.
2. Contour Mapping and Localization: The model should incorporate contour mapping
to create a visual overlay on the medical scans, clearly marking the areas that exhibit
abnormalities, such as lesions or fractures. This helps healthcare workers pinpoint
problematic regions and make more informed decisions.
3. Explainable Reasoning: The AI should not only provide a diagnostic output but also
explain the reasoning behind its conclusions. This can include highlighting key areas in
the image that influenced its decision, such as the size and shape of a lesion or
fracture. Such explainability will be crucial for healthcare workers in understanding
and validating the model’s findings, especially when they lack advanced radiology
training.
4. Resolution and Image Quality: The model must be robust to variations in image
quality and resolution. Medical scans from different machines or in different settings
can vary widely, so the system must be capable of working with low-resolution images
or scans with limited clarity without sacrificing diagnostic performance. Additionally,
it should be able to handle noise and artifacts in the images that may be common in
low-resource environments.
5. Scalability and Efficiency: The system must operate efficiently even with limited
computational resources, such as those found in rural hospitals or clinics with limited
 access to high-performance servers. Fast processing times are essential to ensure that
healthcare workers receive timely assessments, especially in emergency situations.
6. Generalizability: The AI model must be designed to generalize across different patient
populations, imaging devices, and clinical conditions. It should work across diverse
demographic groups, including various ages, genders, and ethnicities, and remain
effective in different healthcare environments with varying levels of imaging
equipment quality.
7. Usability in Resource-Constrained Settings: The interface should be user-friendly,
providing clear visual feedback on the analysis results, such as the contour maps and
highlighted areas of concern. This is particularly important in environments where
healthcare workers may have limited radiological expertise and may need intuitive
tools to guide their decision-making.

The AI system should be designed to help healthcare professionals catch diseases like
tuberculosis, pneumonia, and bone fractures earlier, so they can make quicker, more
informed decisions, especially in places where resources are limited. By using contour
mapping, the AI will clearly highlight areas of concern on the medical images, making it easier
for doctors to spot potential problems. What's more, the system will explain its reasoning
behind each diagnosis, helping healthcare workers understand why certain areas are flagged,
even if they aren't radiology experts. The goal is to make sure the AI works well, even with
low-quality or unclear images, which is often the case in resource-constrained settings.
Ultimately, this tool could help save lives by enabling earlier diagnosis and treatment in areas
where access to expert care is limited.

Domain
Healthcare
Problem Statement:
Personal Finance Planning System to organize, analyze, improve and plan spending,
investments and overall personal finance.

Description:
Many young professionals and college graduates struggle to get a clear, actionable plan for
their finances: tracking bank statement flows, investments, expenses, and mapping them to
goals (like buying a home, saving for a vacation, or reaching a net-worth target). Manual
spreadsheets or generic budgeting apps often fall short of providing personalized guidance,
dynamic adjustments, and integrated goal-tracking.
Design an AI/ML driven system that takes in a user’s financial data (bank statements,
investment holdings, saving goals, recurring expenses) and delivers a comprehensive, goal-
driven personal finance plan, spending habits, analytics, alerts, and intelligent
recommendations.
The goal of solution is to allow user to:
• Understand, analyze and improve their financial wellbeing.
• Analyze their investments
• Allow them to organize and plan their future goals.
Multi-agent system & MCP integration would be considered a plus.

Domain
Banking and Finance
Problem Statement:
An AI Tool to Compare Financial Product Claims Against Actual Customer Experience

Description
Financial products—from fixed deposits and mutual funds to insurance policies—are often
sold based on marketing promises that do not fully align with the complex, legalistic terms in
the fine print. This gap between "Expectation" (marketing/sales pitch) and "Reality" (product
performance/customer experience) is known as mis-selling.
Regulators (Supervisory Authorities) need tools to monitor this gap in real-time, but they are
often reactive, dealing with complaints after the damage is done.
Your team must build a proactive prototype that acts as an independent guardian for financial
consumers. This tool must compare the explicit promises of a product with the lived
experience of the consumers to automatically flag potential mis-selling risks.
The solution must be "decentralized" in its data sources, relying on public and consumer-
generated data rather than direct bank feeds.
The prototype should integrate two main data sources and analytical components:
• The "Expectation" Engine (NLP)
o Input: Simulated Public Documentation (e.g., PDFs of a Mutual Fund Scheme
Information Document, or simulated marketing brochures).
o Analysis: Use Natural Language Processing (NLP) to extract quantifiable promises and
key terms, such as:
▪ Investment Objective/Return Goal (e.g., "Aim to beat the benchmark by
5% over 5 years").
▪ Lock-in Periods, Fees, and Exit Loads.
▪ Risk Profile (e.g., "Low Risk," "Capital Protection").
o Output: A structured digital "Promise Profile" for the product.
• The "Reality" Engine (Sentiment Analysis)
o Input: Simulated large volumes of unstructured consumer feedback (e.g., anonymized
social media posts, online reviews, or simulated complaint logs).
o Analysis: Use Sentiment Analysis and Topic Modelling to track the tone and themes
of customer discussions related to the product or the provider.
o Output: A "Customer Dissatisfaction Index" or a rolling sentiment score, along with a
breakdown of the top 3 negative topics (e.g., "High fees," "Poor service," "Hidden
charges").
Domain
Banking and Finance
Problem Statement:
Develop a real-time machine learning system that analyzes transaction patterns to detect
fraud, using behavioral analytics and anomaly detection to minimize false positives while
identifying genuine suspicious activities.

Description
As digital transactions in the Banking & Finance sector continue to increase, the risk of fraud
has become a significant challenge. Traditional fraud detection systems, which rely on static
rules or post-transaction analysis, often lead to delayed responses, inaccurate alerts, and
overburdened security teams. Additionally, these systems frequently generate high numbers
of false positives, frustrating legitimate customers and disrupting operational workflows.
The goal is to develop a machine learning-based solution that can detect fraud in real-time by
analysing transaction patterns as they happen. By utilizing behavioural analytics, the system
will understand and learn individual user behaviours, enabling it to flag deviations from
normal transaction activities. This will allow for more accurate detection of fraudulent
behaviour while reducing the number of false alarms.
The system will also leverage anomaly detection techniques to identify both known and
novel types of fraudulent activity. It will be capable of detecting unusual behaviours, such as
large transactions, unfamiliar locations, or atypical transaction timings, which may indicate
fraud.

Key objectives include:

1. Real-Time Detection: The system must analyse transactions instantly, providing

immediate alerts for fraudulent activities and minimizing the window of time for
potential fraud to go undetected.

2. Behavioural Analytics: By learning each user’s typical transaction patterns, the system
can accurately identify anomalies and flag suspicious activities with greater precision.

3. Minimizing False Positives: To enhance user experience and operational efficiency,

the system should strike a balance between being sensitive enough to catch fraud
while minimizing false alarms that could overwhelm security teams or inconvenience
customers.

4. Anomaly Detection: The model should not only identify known fraud patterns but also
be able to detect new, evolving fraud tactics that may not conform to traditional
methods.

5. Seamless Integration with Banking & Finance Systems: The solution should integrate
smoothly with existing Banking & Finance infrastructures, ensuring that it works
 alongside current systems without causing major disruptions or requiring extensive
changes.

To take it a step further, the system could incorporate adaptive learning, where it
continuously improves its detection capabilities by learning from new transaction data over
time. Additionally, a user-friendly dashboard could be added, providing real-time insights,
risk scoring, and detailed analytics for fraud prevention teams. This would allow for proactive
decision-making and offer more actionable intelligence, making the system even more
powerful and valuable in a real-world, high-stakes environment.

Domain
Banking & Finance
Problem Statement:
AI-Powered Meeting Summarizer with Task Extraction and Smart Scheduling

Description
In modern work environments, meetings generate significant amounts of information,
including decisions, action items, and follow-ups. However, manually sifting through meeting
recordings or transcripts to capture these key takeaways can be time-consuming and prone
to error. Existing tools often provide basic summaries but fail to automatically extract action
items, assign tasks, and integrate with project management systems like JIRA or Trello.

This solution aims to automate this entire process by analysing meeting recordings or
transcripts, summarizing key discussions, and extracting actionable items. The system will
assign tasks to the appropriate team members and automatically generate tickets/work
items, ensuring that follow-ups are well-organized and deadlines are met. It will also integrate
seamlessly with calendar apps to create reminders for each task, ensuring nothing slips
through the cracks.

A standout feature will be contextual sentiment analysis, which analyses the tone and
urgency of action items, enabling the system to prioritize tasks based on how strongly they
were emphasized during the meeting. This ensures that critical tasks are flagged for
immediate attention, while less urgent ones are scheduled accordingly.

Some challenges that can be addressed include (good to have):

1. Noisy or Incomplete Audio: The system must handle low-quality or partial recordings,
ensuring accurate transcription and content extraction even with poor audio quality
or background noise.

2. Speaker Identification: Accurately identifying and attributing comments, action

items, and decisions to the correct speakers during meetings.

3. Scaling Across Teams & Projects: The platform must scale efficiently across large
organizations with multiple teams, ensuring it can handle a high volume of tasks and
projects without performance degradation.

4. Integration with Diverse Tools: The system must seamlessly integrate with various
project management tools (like JIRA, Asana, Trello) and calendar apps (Google
Calendar, Outlook, etc.), syncing action items, tickets, and reminders without
requiring manual intervention.
 5. Real-Time Processing: The system should generate summaries and task assignments
in near real-time, providing immediate outputs to streamline workflow and decision-
making.

6. Context-Aware Prioritization: Analyzing the urgency of action items based on

emotional and tonal cues from the meeting, allowing the system to prioritize critical
tasks and schedule less pressing items for later.

7. Scheduling Daily Sync-Ups Without Overwhelming: The system should intelligently

schedule daily or weekly sync-ups based on ongoing tasks and team availability,
ensuring that team members are not overwhelmed by frequent, unnecessary
meetings. It should consider the urgency of tasks, existing workloads, and team
members’ availability to avoid scheduling conflicts or burnout.

Domain
Smart Automation
Problem Statement:
Intelligent HR Onboarding System

Description
In today’s fast-paced corporate world, efficient and engaging HR onboarding is crucial for
creating a positive first impression and fostering long-term employee success. However,
many organizations still rely on manual, repetitive processes for onboarding new
employees, which can lead to inefficiencies, delays, and errors. These traditional tasks—
such as document collection, training scheduling, IT account setup, and role assignment—
are time-consuming for HR teams and often result in a fragmented onboarding experience
for new hires. As companies scale and the number of new hires grows, these inefficiencies
become even more pronounced, affecting productivity, engagement, and employee
satisfaction.

To address these challenges, the goal is to develop an Intelligent HR Onboarding

Automation tool that automates the entire onboarding workflow. By integrating
automation, machine learning (ML), and artificial intelligence (AI), this tool will streamline
onboarding tasks, reduce human error, and provide a personalized experience for each new
hire. The solution will focus on automating document checks, task assignments, IT account
creation, and role-based training. This will not only enhance operational efficiency for HR
teams but also improve the overall onboarding experience, ensuring that new employees
are welcomed and integrated quickly and smoothly.

Automated Document Checks

One of the most repetitive and error-prone tasks in HR onboarding is the verification of
documents, such as personal identification, contracts, and tax forms. These documents are
crucial for compliance, yet manually reviewing them can be slow and subject to human
oversight. With Intelligent HR Onboarding Automation, document verification is automated
through the use of Optical Character Recognition (OCR) and AI-driven validation tools. These
technologies allow the system to process and validate documents in real-time, flagging any
discrepancies or missing information. By automating this process, HR teams can save
valuable time, reduce errors, and ensure that employees’ documents meet legal and
organizational standards.

Automated Task and Workflow Assignment

Onboarding involves collaboration across multiple departments, such as IT for account
creation, payroll for salary setup, and facilities for assigning equipment. However, tracking
and managing these tasks manually can lead to confusion and delays. The onboarding tool
solves this problem by automating task assignments based on the employee’s role and
department. Once an employee’s details are entered, the system automatically triggers
relevant tasks—whether it’s creating an IT account, scheduling training, or arranging
equipment. Additionally, notifications are sent to HR and relevant departments, ensuring
that all steps are completed on time and no tasks are overlooked. This workflow automation
improves coordination between departments and guarantees that the onboarding process
progresses smoothly.

Automated IT Account Creation

One of the most tedious tasks in the onboarding process is the creation of IT accounts. IT
teams are often tasked with setting up email accounts, access to internal systems, and
permissions for new employees, which can lead to delays and mistakes if done manually.
The proposed solution integrates with enterprise IT infrastructure through APIs, automating
the creation of accounts based on the employee’s role. The system will generate user
accounts for tools such as email, document storage, internal communication platforms, and
other essential software. By automating these tasks, IT teams are freed up to focus on more
critical tasks, and new employees can begin using necessary tools without unnecessary
delays.

Personalized Onboarding Experience Using Machine Learning

Onboarding is not a one-size-fits-all process. Employees in different roles or departments
require tailored training, resources, and access to information. Traditional onboarding
methods may fail to address these specific needs, leading to disengagement and
inefficiency. Machine learning can address this issue by predicting the new hire’s role and
tailoring their onboarding experience accordingly. The system analyzes historical data, job
descriptions, and skills to recommend role-specific training, documents, and resources. For
example, an employee in a technical role may be directed to specific software training,
while someone in HR may be given access to policy manuals and compliance training. This
personalized approach enhances engagement, accelerates the learning process, and ensures
that employees are equipped with the right tools and knowledge from the outset.

Analytics and Continuous Improvement

Monitoring the effectiveness of the onboarding process is often overlooked, but it is
essential for continuous improvement. By integrating analytics into the onboarding system,
HR teams can track key metrics such as time-to-completion, task completion rates, and
employee satisfaction. This data provides valuable insights into how well the onboarding
process is functioning and where improvements can be made. For instance, if a particular
training module consistently receives negative feedback, HR can make adjustments to
improve its content or delivery. By regularly analyzing onboarding data, organizations can
ensure that the process evolves over time to meet both the needs of employees and the
organization.

Data Privacy and Compliance is a critical concern. Onboarding involves the handling of
sensitive personal data, and companies must ensure that their processes comply with data
protection regulations like GDPR and HIPAA. The solution will incorporate strong encryption
and access controls, ensuring that personal information is securely stored and processed.
Additionally, audit trails will be built into the system to provide transparency and
accountability.

Integration with Existing Systems poses a challenge for companies with legacy systems. The
onboarding tool will need to be flexible, offering open APIs and integration capabilities to
connect seamlessly with existing HRIS, payroll, and IT systems.

Lastly, Employee Engagement during the onboarding process, especially in remote or hybrid
environments, can be challenging. To address this, the solution will include virtual
onboarding features like video training, virtual team introductions, and digital mentorship
programs to ensure that remote employees feel connected and supported.

Domain
Smart Automation