Built PRism, a full-stack GitHub pull request review agent: submit a PR URL, fetch metadata and unified diff, and run a LangGraph workflow for security, correctness, and maintainability feedback. Shipped a FastAPI service that parses diffs into per-file hunks, calls OpenRouter via LangChain, and posts the generated Markdown review back to the PR as a comment. Delivered a Next.js dashboard with Better Auth and Prisma on PostgreSQL for accounts and review history, plus a responsive Tailwind & Shadcn UI with streamed Markdown results.

Built a command-line RAG tool to ingest local files or folders (PDF, TXT, and more) and ask natural-language questions against that corpus in a REPL-style session. Implemented chunking with overlap, embeddings via sentence-transformers, and persisted retrieval in ChromaDB; top-k context is injected through LangChain dynamic_prompt middleware wired to a small agent for streaming answers. Integrated Unstructured for document loading and a Hugging Face–hosted Gemma chat model, with a Click-based CLI and local vector-store persistence for repeat queries.

Built DSA Helper, a full-stack Next.js App Router chat app that streams replies through OpenRouter and the Vercel AI SDK, with Prisma on PostgreSQL for users, sessions, and conversation/message history. Integrated Better Auth (email/password and optional GitHub OAuth) and LangChain-powered RAG over in-repo algorithm study notes, plus a LeetCode problem tool so answers stay grounded in structured context. Shipped a streaming chat UI with AI-generated conversation titles, versioned system prompts, and a Tailwind CSS front end on React 19.

Built a real-time quiz platform for live interactive sessions—hosts create quizzes, participants join via code, responses flow in live. Implemented multiple question types (MCQs, polls, word clouds), timed questions, and rich analytics with WebSockets for sub-50ms response times. Designed responsive UI with Tailwind CSS & Shadcn UI; auth via Clerk, backend on Express + Bun, PostgreSQL for persistence.

Developed a full-stack Next.js app with TypeScript & MongoDB, centralizing Codeforces, LeetCode & CodeChef contests with real-time updates. Designed a responsive UI using Tailwind CSS & Shadcn UI for seamless multi-device support. Added platform filtering, contest bookmarking, YouTube solutions, & theme switching.

Integrated with the Gemini API, it provides advanced expense tracking, predictive analysis, and comprehensive financial insights. Implementation of sophisticated expense categorization and machine learning-powered prediction algorithms offering users actionable financial recommendations. Empowering users with deep, data-driven insights into their spending habits and financial health.