Personal Memory Search Engine

A MINOR PROJECT REPORT

Submitted by
Moin Khatri
Vivek Jangid
Het Patel
220203100046
2202031000038
2202031000094

BACHELOR OF TECHNOLOGY
in
Information Technology

College of Technology

Silver Oak University, Ahmedabad

October, 2025
 College of Technology
Opp. Bhagwat Vidhyapith, S.G. Highway, Ahmedabad-382481

CERTIFICATE
This is to certify that the Internship/Minor Project report submitted along with the
Internship/Minor Project entitled Personal Memory Search Engine has been carried out by
Moin Khatri, Vivek Jangid, Het Patel under my guidance in partial fulfillment for the Bachelor
of Technology in Information Technology, 7th Semester of Silver Oak University, Ahmedabad
during the academic year 2025-26.

Prof. Kamran Saiyyad <Name of Head of the Department>

Internal Guide Head of the Department
 College of Technology
Opp. Bhagwat Vidhyapith, S.G. Highway, Ahmedabad-382481

DECLARATION
We hereby declare that the Internship report submitted along with the Minor Project entitled
Personal Memory Search Engine submitted in partial fulfillment for the Bachelor of
Technology in Information Technology to Silver Oak University, Ahmedabad, is a bonafide
record of original project work carried out by us at College of Technology under the supervision
of Prof. Kamran Saiyyad and that no part of this report has been directly copied from any
students’ reports or taken from any other source, without providing due reference.

Moin Khatri
Vivek Jangid
Het Patel Sign of Student
 List of Figures

Fig 1.1 Use case / Procedure Diagram 2

Fig 1.2 Activity / Process Diagram 5
Fig 2.1 Use case / Procedure Diagram 12
Fig 2.2 Activity / Process Diagram 15
Fig 2.3 Use case / Procedure Diagram 22
Fig 2.4 Activity / Process Diagram 25
Fig 3.1 Use case / Procedure Diagram 32
Fig 4.1 Activity / Process Diagram 35
Fig 4.2 Use case / Procedure Diagram 42
Fig 4.3 Activity / Process Diagram 43
 List of Tables

Table 1.1 Popular Methods / Techniques 2

Table 1.2 User / Reading / Observation Table 5
Table 2.1 Popular Methods / Techniques 12
Table 2.2 User / Reading / Observation Table 15
Table 2.3 Popular Methods / Techniques 22
Table 2.4 User / Reading / Observation Table 25
Table 3.1 Popular Methods / Techniques 32
Table 4.1 User / Reading / Observation Table 35
Table 4.2 Popular Methods / Techniques 42
Table 4.3 User / Reading / Observation Table 43
 Appendix 8

Table of Contents
Acknowledgement i
Abstract ii
List of Figures iii
List of Tables iv
List of Abbreviations v
Table of Contents vi
Chapter 1 Introduction 1
1.1 Introduction to Topic 2
1.1.1 How to do 3
1.1.1 How to do 3
1.1.1 How to do 4
1.2 Introduction to Topic 4
1.3 Introduction to Topic 5
1.4 Introduction to Topic 6
Chapter 2 Title of chapter 8
2.1 Introduction to Topic 9
2.1.1 How to do 10
Chapter 3 Title of chapter 11
3.1 Introduction to Topic 14
3.1.1 How to do 16
3.2 Introduction to Topic 17
References 40
Appendix. 42
 CHAPTER 1: INTRODUCTION

1.0 OVERVIEW OF THE COMPANY

The project Personal Memory Search with Voice and Image has been developed under the
Department of Information Technology at Aditya Silver Oak Institute of Technology.
The project team functions like a small software development company with dedicated roles in
design, coding, testing, and documentation. The objective of the team is to build an innovative AI-
based application that enhances digital memory management through intelligent search methods.
The team follows a professional software development process including planning, designing, and
implementation using modern technologies like Python, Streamlit, and AI models.

1.1 HISTORY
With the rapid advancement of artificial intelligence and cloud computing, applications capable of
understanding human language, voice, and visual content have become increasingly popular.
Inspired by digital assistants such as Google Lens and ChatGPT, the project team initiated this
development in 2025.
The project evolved from the idea of creating a personal tool where users can upload their text,
images, or voice notes and retrieve them later through simple queries. Over time, the team enhanced
the system by integrating NLP, speech recognition, and image analysis models to achieve higher
accuracy and convenience.

1.2 DIFFERENT PRODUCTS / SCOPE OF WORK

The project focuses on the creation of a personal data management system.
The scope of work includes:
 Developing a user-friendly web-based application using Streamlit.
 Allowing users to store memories (text, voice, image).
 Implementing semantic search using AI models.
 Providing efficient retrieval of stored data through natural queries.
 Ensuring data privacy and smooth performance for personal use.
In broader scope, the same technology can be extended to note-taking apps, office assistants, or
digital diaries.

1.3 ORGANIZATION CHART

The development team works in a structured manner, similar to a professional organization.
Project Guide: Mr. Kamran Saiyyad
Team Members:
 Moin Khatri (2202031000046)
 Vivek Jangid (2202031000046)
 Het Patel (2202031000094)
Roles:
 Team Lead: Coordinates tasks, manages timelines, and ensures integration.
 Developer: Implements coding, testing, and debugging.
 Documentation Lead: Prepares reports and maintains records.
1.4 CAPACITY OF PLANT

As the project is based on software development, the “plant capacity” refers to the computational
and development capabilities of the system used.
Development was carried out on personal computing environments with configurations sufficient for
AI and data handling tasks, including:
 Processor: Intel i5 / i7 or equivalent
 Memory: 16 GB RAM
 Storage: 512 GB SSD
 Software: Python 3.10, Streamlit, SQLite
 Operating System: Windows 11
The system is capable of running real-time searches for small to medium-sized datasets and can be
scaled up further using cloud services.
CHAPTER 2: OVERVIEW OF SYSTEM COMPONENTS AND
PROCESS LAYOUT

2.0 OVERVIEW OF DIFFERENT MODULES

The application functions as a compact AI-based system made up of interconnected modules that
handle data input, processing, storage, and retrieval. Each module plays a key role in transforming
user input—text, voice, or image—into searchable information.
The main modules are:
1. User Interface Module – Provides a simple Streamlit web interface for uploading data and
performing searches.
2. Speech Recognition Module – Converts spoken input into text using Whisper or
SpeechRecognition libraries.
3. Image Processing Module – Extracts visual features and captions from uploaded images.
4. Natural Language Processing Module – Generates text embeddings and performs semantic
comparison.
5. Database Module – Stores processed data (text, voice transcripts, image details) in SQLite
for efficient retrieval.
6. Search and Retrieval Module – Matches user queries with stored embeddings and displays
ranked results.

2.1 Work Carried Out in Each Module

 User Interface: Designed using Streamlit to ensure minimal coding complexity and rapid
interaction. Users can easily upload files and view results instantly.
 Speech Recognition: Audio inputs are cleaned, converted to text, and sent to the NLP
module for embedding generation.
 Image Module: Uses libraries like Pillow and OpenAI Vision models to analyze uploaded
images and describe their content in text form.
 NLP Module: Employs transformer-based models to create embeddings for semantic
understanding.
 Database Module: Handles storage and retrieval operations, linking each file with metadata.
 Search Module: Implements cosine similarity to fetch the most relevant results to the user’s
query.

2.2 Technical Specifications of Major Components

Component Specification / Library Used
Programming Language Python 3.10
Framework Streamlit
Speech Recognition Whisper / SpeechRecognition
Image Handling Pillow, OpenAI Vision API
NLP & Search Sentence Transformers / OpenAI Embeddings
Database SQLite
Front-End Design Streamlit widgets and layout
Hardware Requirement Minimum 8 GB RAM, i5 Processor
Operating System Windows 11 or equivalent Linux system
CHAPTER 3: INTRODUCTION TO PROJECT / PROJECT
MANAGEMENT

3.0 INTRODUCTION TO THE PROJECT

The project Personal Memory Search with Voice and Image is an AI-driven application that allows
users to store and retrieve memories using natural inputs such as text, voice, and images. It provides
a smart, human-like way of accessing personal information without manually browsing through
folders or filenames.
The idea was inspired by the increasing need for intelligent personal assistants that can organize and
recall data naturally, enhancing productivity and convenience in everyday digital life.

3.1 PROJECT SUMMARY

This project integrates speech recognition, image captioning, and natural-language processing within
a Streamlit web application. Users can upload text, image, or audio files that are automatically
processed and stored. When queried later—through either text or voice—the system analyzes the
input and returns the most semantically relevant results.
The core principle of this project is embedding-based semantic similarity, which enables accurate
matching even when the search query does not exactly match stored data.

3.2 PURPOSE
The purpose of this project is to create an intelligent, easy-to-use memory management system that:
 Allows users to save important personal data in multiple formats.
 Enables retrieval through natural-language and voice queries.
 Demonstrates the integration of AI models for everyday problem-solving.
 Reduces the time spent searching for stored information.

3.3 OBJECTIVES
1. To design an AI-powered application capable of handling multimodal input (text, voice,
image).
2. To implement voice-to-text conversion using modern speech-recognition models.
3. To extract key information from images using computer-vision techniques.
4. To store and manage all data efficiently in a local database.
5. To perform accurate search and retrieval using semantic similarity.
6. To provide an intuitive, interactive, and responsive web interface.

3.4 SCOPE
What it can do:
 Accept, process, and store text, voice, and image data.
 Perform intelligent search using natural-language queries.
 Display relevant results instantly through a user-friendly interface.
What it cannot do (current limitations):
 Handle extremely large datasets or cloud-based storage.
 Recognize complex real-world images with multiple overlapping objects.
 Perform multi-user authentication or role-based access.
Future versions can overcome these limits by integrating cloud storage, improved models, and
authentication systems.
3.5 TECHNOLOGY AND LITERATURE REVIEW
The project builds on existing AI research in three primary domains:
1. Natural Language Processing (NLP):
Utilizes transformer-based models for semantic understanding, inspired by technologies like
BERT and GPT.
2. Speech Recognition:
Uses open-source Whisper or SpeechRecognition APIs for accurate audio-to-text
transcription.
3. Image Understanding:
Applies image captioning and visual embedding models to interpret image content.
Several prior works such as semantic search engines and digital voice assistants formed the
conceptual backbone of this system. However, this project distinguishes itself by combining these
capabilities in a single, lightweight Streamlit application.

3.6 PROJECT PLANNING

3.6.1 Development Approach and Justification
The system was developed using an Agile-iterative approach. Development occurred in short
cycles involving planning, coding, testing, and refining.
This approach ensured rapid prototyping, quick debugging, and flexibility in integrating new models
or features.
3.6.2 Effort, Time, and Cost Estimation
 Time Duration: 3 months (including design, coding, and testing)
 Team Size: 3 members
 Effort Distribution:
o Design & UI – 25 %
o Coding – 45 %
o Testing – 20 %
o Documentation – 10 %
 Cost: Minimal (open-source tools and free APIs used)
3.6.3 Roles and Responsibilities
 Moin Khatri: Team Lead & Developer – coordinated work, handled NLP integration.
 Vivek Jangid: Front-End Designer – implemented Streamlit UI and visuals.
 Het Patel: Data Handling & Testing – managed database and performance validation.
 Mr. Kamran Saiyyad: Project Guide – provided guidance, reviews, and feedback.
3.6.4 Group Dependencies
Team members collaborated using shared GitHub repositories and version control. Tasks were
interdependent—front-end testing relied on back-end integration, and model tuning depended on
accurate data handling.

3.7 PROJECT SCHEDULING

The overall project timeline was structured as follows:
Phase Duration Activities
Requirement Analysis Week 1 – 2 Problem definition, data collection
Design & Planning Week 3 – 4 UI design, architecture layout
Development Week 5 – 8 Module implementation, integration
Testing & Debugging Week 9 – 10 Unit testing, bug fixing
Documentation & Review Week 11 – 12 Report writing, presentation preparation

CHAPTER 4: SYSTEM ANALYSIS

4.0 OVERVIEW
System analysis is a crucial step in software development that involves understanding existing
systems, identifying problems, and defining the requirements for the new system. This chapter
provides a detailed examination of the current challenges in memory management and explains how
the proposed system resolves them using AI-driven techniques.

4.1 STUDY OF CURRENT SYSTEM

In the traditional approach, users manually store and search their personal data across various folders
or devices. Searching through numerous files or images consumes time and often leads to
frustration.
Existing file explorers and gallery apps offer only keyword-based searches, which fail if users don’t
remember exact filenames or folder paths.
There is no unified platform that allows searching across voice, image, and text data using natural
language. This gap forms the basis for developing the Personal Memory Search with Voice and
Image application.

4.2 PROBLEMS AND WEAKNESSES OF THE CURRENT SYSTEM

1. Manual Searching: Users must remember filenames and locations.

2. Limited to One Format: No integration between text, voice, and images.
3. Time-Consuming: Searching large datasets is inefficient.
4. No Semantic Understanding: Keyword-based systems cannot interpret meaning or context.
5. Poor Accessibility: No support for voice-based search or multimodal retrieval.
These limitations highlight the need for a smart, context-aware, and AI-integrated solution.

4.3 REQUIREMENTS OF THE NEW SYSTEM

Functional Requirements

 Ability to upload and store data in text, image, and voice format.
 Convert voice input to text using speech recognition.
 Extract features or captions from images.
 Generate and store embeddings for semantic understanding.
 Allow users to search using natural language or voice queries.
 Display relevant search results instantly.

Non-Functional Requirements

 The interface should be user-friendly and responsive.

 The system should perform searches efficiently even with multiple entries.
 Data storage should be lightweight yet reliable (SQLite).
 The system should maintain security and prevent data corruption.
4.4 SYSTEM FEASIBILITY

4.4.1 Technical Feasibility

The system uses widely available and open-source technologies like Python, Streamlit, and SQLite.
The models for speech and image recognition are accessible via free APIs, making the system
feasible for academic and personal deployment.

4.4.2 Economic Feasibility

No major investment is required as all tools are open-source. The only cost involved is for hardware
resources such as a laptop or PC and internet connectivity for API access.

4.4.3 Operational Feasibility

The system is easy to use with minimal training. Any user can upload files, perform searches, and
retrieve information intuitively. Thus, it is highly feasible for implementation and further expansion.

4.5 ACTIVITY / PROCESS IN NEW SYSTEM

The workflow of the new system can be summarized as:

1. User Input: User provides data (text, voice, or image).
2. Processing: Data is analyzed and converted to a digital representation (text embeddings).
3. Storage: Processed data is stored in a structured SQLite database.
4. Querying: User enters a query through text or voice.
5. Search & Retrieval: Semantic similarity algorithms identify the most relevant matches.
6. Display: Results are shown to the user with corresponding images or notes.

4.6 FEATURES OF THE NEW SYSTEM

 Voice-based input and search functionality.

 Image-to-text conversion and analysis.
 Semantic search using embeddings.
 Easy-to-use Streamlit interface.
 Real-time result display.
 Lightweight and fully local execution.

4.7 MAIN MODULES / COMPONENTS

1. Voice Recognition Module – Converts audio into text using AI models.

2. Image Analysis Module – Generates captions and detects visual elements.
3. Text Embedding Module – Converts text into numerical vectors for similarity search.
4. Database Module – Manages and indexes stored data.
 5. Search Engine Module – Matches queries with stored embeddings using cosine similarity.
6. User Interface Module – Provides a simple front-end for all user interactions.

4.8 SELECTION OF HARDWARE / SOFTWARE / ALGORITHMS /

METHODOLOGY

 Hardware: Intel i5 processor, 8–16 GB RAM, SSD storage.

 Software: Python 3.10, Streamlit, SQLite, Whisper API, Pillow, Transformers.
 Algorithms:
o Speech Recognition – Whisper model.
o Image Captioning – Vision Transformer or OpenAI model.
o Semantic Search – Cosine similarity using vector embeddings.
 Methodology: Modular design with Agile development approach.

Justification:
These technologies and methods were selected because they are lightweight, accurate, and suitable
for academic and small-scale deployment. Streamlit simplifies the UI, and embedding-based search
ensures precise results.

CHAPTER 5: SYSTEM DESIGN

5.1 SYSTEM DESIGN & METHODOLOGY

System design focuses on creating a clear and efficient structure for the Streamlit-based data
visualization application. The main objective is to translate user requirements into a functional
system that allows data upload, processing, and visualization in real time.
The design follows a modular architecture, dividing the system into key components:
 Frontend (UI Layer): Designed using Streamlit’s interactive widgets for user input and
visualization.
 Backend (Processing Layer): Handles dataset loading, data preprocessing, and machine
learning model execution.
 Visualization Layer: Generates dynamic charts and plots using Matplotlib, Plotly, and
Seaborn.
 Storage Layer: Handles temporary in-memory storage of datasets using Pandas
DataFrames.
The methodology adopted for this project is Agile, enabling iterative improvement and user-centered
design through regular feedback and testing.

5.2 DATABASE / DATA STRUCTURE DESIGN

The project doesn’t use an external database; instead, it utilizes Pandas DataFrames for handling
structured datasets dynamically.
Key Data Structures:
 DataFrame: To store uploaded CSV or Excel data.
 Lists and Dictionaries: For temporary storage of metadata such as column names, data
types, and selected features for analysis.
 Model Objects: For storing trained machine learning models (if applicable).
This approach provides high flexibility and eliminates dependency on external databases, making
the application lightweight and fast.

5.3 INPUT / OUTPUT AND INTERFACE DESIGN

5.3.1 Input Design

Inputs are provided by the user through Streamlit’s UI components:
 File uploader (for CSV/Excel datasets)
 Text inputs (for column selection or custom queries)
 Dropdowns and sliders (for choosing visualizations or parameters)
 Buttons (to trigger model training or graph generation)
All inputs are validated for type and format before processing to avoid runtime errors.

5.3.2 Output Design

Outputs are primarily visual and analytical:
 Graphs such as bar charts, line plots, scatter plots, and correlation heatmaps.
 Statistical summaries like mean, median, and standard deviation.
  Model performance metrics (if machine learning is applied).
Outputs are displayed directly on the web interface in real-time after user interaction.

5.3.3 Interface Design

The interface uses a clean, card-based layout with intuitive navigation.
Sections include:
1. Home Page – Overview and file upload option.
2. Data Preview Section – Displays uploaded data in table format.
3. Visualization Section – Generates charts and plots dynamically.
4. Model Section (optional) – Allows training simple ML models on uploaded data.
The UI maintains consistent colors, typography, and spacing for professional readability.

5.3.4 State Transition Diagram

Below is the logical flow of the system:
User Uploads File → Data Validation → Data Display → User Selects Visualization →
Visualization Generated → (Optional) Model Training → Results Displayed

5.3.5 Access Control / Security

Although this is a single-user academic project, basic safeguards are implemented:
 File type validation (accepts only .csv and .xlsx files)
 Error handling for invalid data or missing columns
 Streamlit’s secure session state to handle temporary user interactions safely
If deployed online, HTTPS and user authentication can be added for data privacy and integrity.
CHAPTER 6: IMPLEMENTATION

6.1 IMPLEMENTATION PLATFORM / ENVIRONMENT

The application was implemented using Python as the core programming language and Streamlit as
the web framework.
All development and testing were carried out on the following setup:
 Programming Language: Python 3.12
 Framework: Streamlit
 Libraries Used: Pandas, NumPy, Matplotlib, Seaborn, Plotly, Scikit-learn
 IDE / Editor: Visual Studio Code
 Operating System: Windows 11
 Hardware Used: Lenovo LOQ 15IRX9 (Intel i5-13450HX, RTX 4050, 16GB RAM)
 Version Control: GitHub (for source code backup and version tracking)
Streamlit was chosen for its simplicity in building data applications with minimal code while
providing an interactive and modern web interface.

6.2 PROCESS / PROGRAM / MODULE SPECIFICATION

The system was divided into modules to ensure clarity, maintainability, and scalability.
1. Data Upload Module:
Allows users to upload CSV or Excel datasets using Streamlit’s file uploader. Performs validation
on file format and structure before processing.
2. Data Preprocessing Module:
Handles missing values, detects data types, and generates statistical summaries. Uses Pandas for
operations like cleaning, sorting, and filtering.
3. Visualization Module:
Generates different charts based on user selection, such as:
 Bar Chart
 Line Chart
 Scatter Plot
 Heatmap (for correlation)
Visuals are rendered dynamically with Streamlit and Plotly, enabling real-time updates.
4. Machine Learning Module (optional):
Implements simple models like Linear Regression or Classification (if included in the app) using
Scikit-learn. Displays performance metrics such as accuracy, RMSE, or confusion matrix.
5. Output and Result Display Module:
Displays all generated results and graphs instantly within the Streamlit interface, ensuring an
interactive experience for the end-user.

6.3 FINDINGS / RESULTS / OUTCOMES

After successful implementation, the following outcomes were achieved:

  A fully functional web-based data visualization app using Streamlit.
 Users can upload any dataset and instantly visualize it without coding knowledge.
 The app automatically generates statistical summaries and interactive charts.
 The modular structure ensures easy future enhancements, like adding ML or AI-based
analysis.
 The system runs efficiently with minimal hardware requirements.
The app significantly simplifies exploratory data analysis (EDA) for non-technical users and can be
extended for educational or business data visualization use cases.

6.4 RESULT ANALYSIS / COMPARISON / DELIBERATIONS

The project’s performance was evaluated based on speed, accuracy, and usability.

Criterion Observation Result

Under 2 seconds for 5MB
File Upload & Load Time Excellent
dataset
Visualization Speed Real-time rendering Excellent
Data Handling Smooth up to 100k records Satisfactory
UI/UX Design Intuitive and responsive Excellent
Managed invalid inputs
Error Handling Good
gracefully

CHAPTER 7: TESTING
7.1 Testing Plan / Strategy

Testing was carried out systematically to ensure the accuracy, reliability, and stability of the
Streamlit application. The primary goal was to verify that all modules performed as expected and
that the user interface remained responsive across different datasets.
The testing approach followed a bottom-up strategy, starting from individual components and
progressing toward the integrated system.
Types of Testing Conducted:
1. Unit Testing:
Each function and component (file upload, visualization generation, etc.) was tested
independently.
2. Integration Testing:
Ensured smooth data flow between modules like data upload → preprocessing →
visualization.
3. Functional Testing:
Verified that the system met all specified requirements, such as dynamic chart generation
and dataset summary.
4. User Acceptance Testing (UAT):
Conducted to check usability and interface quality. Real users (project team members)
interacted with the app and provided feedback.
5. Performance Testing:
Measured load time and responsiveness when handling large datasets.

7.2 Test Results and Analysis

The testing confirmed that the application works reliably with no major bugs or performance issues.
The interface handled multiple types of datasets smoothly, and all outputs were generated as
expected.
Summary of Observations:
 The app successfully handled various dataset sizes (from 1MB to 20MB).
 Visualizations updated instantly upon user input.
 Invalid file formats were blocked gracefully with clear error messages.
 Overall, the system passed all test cases with high reliability and accuracy.

7.2.1 Test Cases

Test Test Condition Expected Output Actual Output Remarks
ID
T001 Upload valid CSV file Dataset loads and Works as expected Pass
preview displays
T002 Upload invalid file Error message displayed Error handled Pass
format (.txt) correctly
T003 Select column for Graph generated Works as expected Pass
visualization instantly
T004 Upload large dataset Should load within 5 Loaded in 4.2 seconds Pass
 (20MB) seconds
T005 Missing values in Should handle Missing values Pass
dataset gracefully filled/ignored
T006 Generate correlation Correlation chart Correct and Pass
heatmap displayed responsive
T007 Invalid column Display user-friendly Message displayed Pass
selected error properly
T008 Restart app and re- State resets and reloads Works correctly Pass
upload data cleanly