LuminaQA - RAG

LuminaQA is a smart, Retrieval-Augmented Generation (RAG) system that enables users to ask questions based on document collections (e.g., holy books) and receive intelligent, context-rich responses. Built with LlamaIndex, LangChain, and Groq's LLMs, it offers both LLM-generated and strictly document-sourced answers. Users can interact via a console or a Flask-based web UI. It maintains chat history to provide contextual understanding in both LLM and source-based outputs, improving relevance and coherence.

Key Achievements:

• 🧠 Context-Aware Responses: Maintains chat history to ensure both LLM and document-based answers are contextually accurate.
• 🔄 Smart Re-Indexing: Uses document hashing to re-index only changed files, boosting performance.
• 📂 Multiple Output Modes: Offers LLM-generated answers, source-only responses, document metadata (filename, page), and relevance metrics.
• 🌐 Dual Interface Support: Console and web interfaces provide flexible usage options.
• 📊 Search & Relevance Metrics: Displays how relevant a response is with performance scoring.

Skills & Technologies Used:

• 🧠 AI & NLP: LlamaIndex, LangChain, Groq LLMs
• 💬 Context Handling: Persistent chat history, token-saving summarization
• 🌐 Web Framework: Flask
• 🐍 Backend: Python
• 📊 Embedding Model: Sentence Transformers (all-MiniLM-L6-v2)
• 🛠️ Deployment: Docker, Git, GitHub

RealtimeSTT_Websocket

RealtimeSTT_Websocket is a real-time Speech-to-Text (STT) system that streams audio from the browser via WebSocket for fast, accurate transcription. Ideal for developers integrating instant speech recognition into their applications, this system leverages the Vosk speech recognition model for efficient processing.

Key Achievements:

• 🗣️ Real-Time Transcription: Implemented a system that converts spoken language into written text instantly, enhancing user interaction in applications.
• 🔗 WebSocket Integration: Utilized WebSocket for continuous, low-latency communication between the browser and server, ensuring seamless data transmission.
• 🧠 Efficient Speech Recognition: Employed the Vosk speech recognition model, known for its accuracy and lightweight performance, suitable for real-time applications.
• 🌐 Browser Compatibility: Developed a solution that captures audio from the browser microphone, making it accessible across various platforms without additional plugins.
• 📦 Open-Source Contribution: Released the project under the MIT license on GitHub, providing a valuable resource for developers seeking to implement real-time STT functionality.

Skills & Technologies Used:

• 🧠 Speech Recognition: Vosk Model, OpenAI_Whisper(HuggingFace)
• 🌐 Web Communication: WebSocket
• 💻 Frontend Development: HTML, JavaScript
• 🐍 Backend Development: Python, fastapi
• 📦 Package Management: pip

SmartMedReport_Dashboard

SmartMedReport Dashboard is an AI-powered web application designed to revolutionize the analysis of medical reports and insurance claims. By leveraging Groq's advanced Large Language Models (LLMs), the platform enables healthcare professionals and insurers to:

• Uncover hidden patterns in patient data
• Identify unnecessary or redundant medical tests
• Validate insurance claims against policy documents
• Estimate potential cost savings

The application features a responsive frontend dashboard built with modern web technologies, ensuring an intuitive user experience.

Key Achievements:

• 🤖 AI-Powered Insights: Integrated Groq's LLMs to analyze medical reports and insurance claims, extracting meaningful and actionable insights.
• 📊 Data-Driven Decision Making: Visualized patient journeys, flagged unnecessary procedures, and calculated potential cost reductions for optimized resource usage.
• 🔗 Seamless Integration: Built a full-stack system using FastAPI and modern web technologies to ensure fast, efficient data processing and user interactions.
• 🧑‍💻 User-Centric Design: Created an intuitive and responsive dashboard interface for ease of use by medical and insurance professionals.
• 🌍 Open-Source Contribution: Published on GitHub to support community learning and collaboration in AI-driven healthcare tech.

Skills & Technologies Used:

• 🧠 AI & NLP: Groq's Large Language Models (LLMs)
• 🛠️ Backend Development: FastAPI
• 💻 Frontend Development: HTML, CSS, JavaScript
• 🐙 Version Control: Git & GitHub
• 📦 Deployment: Docker

DeepSudo_Solver

DeepSudo_Solver is an innovative application that leverages deep learning and computer vision to solve Sudoku puzzles directly from images. Users can upload a photo, input a CSV file, or manually enter a puzzle. The system employs a Convolutional Neural Network (CNN) for digit recognition, OpenCV for image processing, and a classical backtracking algorithm to compute the solution. The user-friendly interface is built with Streamlit, making it accessible and interactive.

Key Achievements:

• 🧠 Built a CNN model using TensorFlow to recognize handwritten digits from Sudoku puzzles with high accuracy.
• 🖼️ Automated image processing pipeline using OpenCV to detect and extract Sudoku grids from raw images.
• 🧮 Implemented a classic backtracking algorithm to solve puzzles reliably and efficiently.
• 🔁 Enabled multi-input functionality (image upload, CSV, manual entry) for flexible puzzle input.
• 🌐 Developed an interactive UI using Streamlit to make the tool accessible and easy to use.
• 💻 Open-source release on GitHub, contributing to the AI and computer vision community.

Skills & Technologies Used:

• Deep Learning: TensorFlow, Convolutional Neural Networks (CNN)
• Programming Language: Python
• Web Framework: Streamlit
• Computer Vision: OpenCV
• Algorithms: Backtracking (for puzzle solving)
• Data Handling: CSV, image preprocessing

Spam Message Detection

In this comprehensive end-to-end project, I seamlessly integrated multiple machine learning techniques to tackle the intricate challenge of spam message detection. Leveraging advanced methods in natural language processing with NLTK and incorporating CountVectorizer for feature extraction, I meticulously honed the model's capabilities.

Key Achievements:

• Diverse ML Techniques: Applied a range of machine learning algorithms, ensuring a thorough exploration of models to capture nuances in spam detection.
• Hyperparameter Tuning: Employed rigorous hyperparameter tuning to optimize model performance, achieving a balance between precision and recall.
• NLP with NLTK: Utilized Natural Language Processing techniques with the NLTK library to enhance the model's understanding of contextual information within messages.
• Efficient Feature Extraction: Implemented CountVectorizer for efficient feature extraction, contributing to the model's ability to discern patterns and characteristics of spam messages.
• Streamlit Deployment: Streamlined the user experience by deploying the model through Streamlit, creating a user-friendly interface for real-time interaction and testing.

This project not only demonstrates my proficiency in machine learning and natural language processing but also underscores my commitment to delivering practical, deployable solutions for real-world challenges.

Women's Clothing and Ecommerce Analysis Project

Conducted an in-depth analysis of Women's Clothing and Ecommerce data, employing advanced data analysis techniques using Python libraries including NumPy, Pandas, and Matplotlib. This project aimed to derive meaningful insights into customer behavior, trends, and overall business performance within the realm of women's clothing ecommerce.

Key Components and Achievements:

1. Data Exploration and Cleaning:

• Conducted comprehensive data exploration to understand the structure and nature of the dataset.
• Employed data cleaning techniques using Pandas to ensure data integrity and eliminate inconsistencies.

2. Customer Behavior Analysis:

• Utilized NumPy for statistical analysis to uncover patterns in customer behavior, including preferences, purchase trends, and popular categories.
• Generated visual representations with Matplotlib to effectively communicate insights.

3. Sales and Revenue Trends:

• Conducted a deep dive into sales data, identifying peak sales periods and revenue-driving factors.
• Implemented Pandas for data aggregation and analysis, providing a clear overview of sales trends over time.

4. Market Basket Analysis:

• Leveraged advanced analytics techniques to perform market basket analysis, revealing product associations and aiding in strategic product placement.

5. Outcome and Recommendations:

• Derived actionable insights to enhance the ecommerce platform's performance and user experience.
• Presented a detailed report outlining key findings, trends, and strategic recommendations based on the analysis.

Letter of Recommendation: Received a Letter of Recommendation attesting to the depth of analysis, problem-solving skills, and valuable contributions made to the project.

Conclusion: This project not only showcases proficiency in data analysis and visualization but also demonstrates the ability to extract meaningful insights to drive strategic decision-making in the realm of ecommerce, particularly within the Women's Clothing sector

Storytelling Project: Exploring Profitability in Superstore Using Tableau

Problem Statement:

Investigating the Impact of Sales Increase on Profit

Key Insights:

1. Sales vs. Profit Analysis:

• Examined the correlation between sales and profit to understand the direct impact on profitability.
• Identified regions where increasing sales did not necessarily translate to increased profit.

2. Top Tables Sales, Yet Highest Loss:

• Surprising discovery: Tables rank 4th in sales but incur the highest losses.
• Addressed the challenge of maximizing profit in a product category with significant sales.

3. Profit-Loss Trend Over Time:

• Uncovered a concerning trend: Losses consistently increased over time, reaching a peak in 2018.
• Explored the factors contributing to this trend and formulated strategies for reversal.

4. Regional Profit Variations:

• Regional analysis highlighted Washington and Virginia as states consistently generating the most profit.
• Contrasted this success with the struggle of tables in New York, indicating potential areas for improvement.

5. Discount Sensitivity of Tables:

• Identified sensitivity to discounts as a critical factor contributing to table-related losses.
• Proposed a strategy to cease or strategically modify discounts on tables to boost profitability.

Conclusion: This Tableau-driven storytelling project not only unveils critical insights into the Superstore's sales and profit dynamics but also provides actionable recommendations. By addressing specific challenges, such as regional variations and product sensitivity to discounts, we aim to optimize profitability and guide strategic decision-making for sustainable business growth.