Projects | Projects

ClearBill.AI: Explaining Medical Bills with AI and RAG July 2025 Completed

GitHub Docs

An AI-powered chatbot that uses Retrieval-Augmented Generation (RAG) with Astra DB, LangChain, and Hugging Face’s Llama-3.1-8B-Instruct to help users understand medical bills with clear, context-aware responses.

Built with Next.js and TypeScript, featuring a sleek, responsive chat UI for seamless user interaction
Retrieves answers using semantic search over a vector database populated with scraped healthcare resources
Uses @xenova/transformers for local embedding and LangChain for intelligent text processing
Runs LLM inference with Llama-3.1-8B-Instruct via Hugging Face API with context injection
Includes a custom script to scrape, embed, and load structured healthcare content into Astra DB
CI/CD pipeline using GitHub Actions for automatic linting and builds on push

RAG LLM LangChain Next.js TypeScript Vector Search Astra DB Healthcare AI Web Scraping Puppeteer Hugging Face Mistral Llama-3

SentiSync: Real-Time YouTube Sentiment Analysis with MLOps July 2025 Completed

GitHub Docs

A full MLOps pipeline and Chrome extension for real-time sentiment analysis on YouTube comments, using Flask, MLflow, Docker, and AWS.

Chrome Extension scrapes YouTube comments and fetches real-time sentiment insights from the backend
End-to-end MLOps workflow using DVC for reproducibility and MLflow for experiment tracking
Flask backend deployed via Docker on AWS EC2, with GitHub Actions-based CI/CD pipeline
Supports sentiment prediction, word cloud generation, and trend visualizations through API endpoints
Includes a suite of Jupyter notebooks for model experimentation, evaluation, and ensemble learning

MLOps MLflow DVC Docker AWS Flask Chrome Extension Sentiment Analysis NLP LightGBM

IntelliCrawl: Autonomous AI Research Agent for Developer Tools July 2025 Completed

GitHub Docs

An advanced autonomous research agent that uses LLMs and web crawling to analyze and compare developer tools. Built with LangGraph, LangChain, and Firecrawl for scalable, multi-step analysis workflows.

Supports batch, single, and interactive CLI modes with customizable output (text, markdown, JSON, CSV)
Uses LangGraph for async, multi-step workflows and LangChain for LLM orchestration
Scrapes and analyzes developer tool websites using Firecrawl and GPT-4o-mini
Implements persistent caching with diskcache and .env-configured API keys
Includes a secondary lightweight agent using Firecrawl MCP and GPT-4.1-nano

LLM LangChain LangGraph OpenAI Firecrawl CLI AI Agents

LaTeX Resume Template July 2025 Completed

GitHub Docs

A clean, ATS-friendly, and modern LaTeX resume template built for students and developers applying to technical roles. Features minimalist design, modular macros, and PDF outputs optimized for readability and machine parsing.

Modular LaTeX macros for easy customization and consistency across sections
Single-page layout with bolded technologies and quantifiable bullet points
Fully ATS-compatible PDF output using Unicode embedding and semantic structure
Includes Overleaf support, documentation, and contribution guidelines
Open-source under MIT License, ready to fork and adapt

LaTeX Resume ATS PDF Open Source

FaceTrack: Smart Face Recognition Attendance System July 2025 Completed

GitHub

A full-stack web application that automates attendance tracking using facial recognition. Built with FastAPI, React, and SQLite, it supports real-time webcam input and batch image uploads for seamless attendance logging.

Real-time face recognition via webcam and batch image uploads
FastAPI backend with REST API endpoints and SQLite integration
React frontend with live feedback, image previews, and attendance viewer
OpenCV and face_recognition for robust face encoding and matching
Interactive database-backed attendance logs with export support

Computer Vision OpenCV FastAPI React face_recognition SQLite

CodePrep.AI: AI-Powered Coding Interview Prep with LLMs July 2025 Completed

GitHub Architecture

An interactive full-stack web application that generates multiple-choice coding challenges using Meta-Llama-3-8B-Instruct. Designed for interview practice, daily challenge tracking, and progress feedback with real-time explanations.

Dynamic challenge generation using Meta-Llama-3-8B-Instruct from Hugging Face
Clerk authentication with quota tracking and personalized history view
Full-stack system with React (Vite), FastAPI, and SQLite
Support for challenge explanation, user feedback, and difficulty selection

LLM AI Interview Prep FastAPI React Hugging Face SQLite

DDPG: Deep Deterministic Policy Gradient Reimplementation in TensorFlow 2.x July 2025 Completed

GitHub Architecture Diagram Original Paper

A robust, modular, and extensible reimplementation of the DDPG reinforcement learning algorithm (Lillicrap et al., 2015) using TensorFlow 2.x. Built for reproducibility, research comparison, and interactive experimentation with Gym environments.

Modular architecture: separate agent, networks, noise, buffer, wrappers
Hyperparameter sweeps, YAML config system, and TensorBoard logging
Includes full unit test suite and reproducibility controls (seed, version logging)
Supports Gym ≥ 0.26, Apple Silicon acceleration (tensorflow-macos, -metal)

Reinforcement Learning DDPG TensorFlow OpenAI Gym

ACME10-HE-RAGApp: RAG with Vector Search, Knowledge Graphs, and LLMs May 2025 Completed

GitHub Final Report

A full-stack Retrieval-Augmented Generation (RAG) application developed for HackerEarth that integrates vector search (FAISS), knowledge graphs (DBpedia), and OpenAI’s LLM to generate traceable, context-rich answers from a Wikipedia-based knowledge base.

Developed for HackerEarth using 10,000+ Wikipedia articles
FAISS-based vector search for semantic retrieval
Knowledge graph context enhancement with DBpedia and SPARQL
OpenAI LLM for coherent and natural language answers
Dockerized deployment with local dev option

LLM RAG Vector Search Knowledge Graph Full-Stack NLP

Project Gallery