Built a visual workflow automation platform with intuitive drag-and-link interface for building, executing and monitoring workflows via triggers and actions. Turborepo architecture with front-end using React, TypeScript, Vite and Tailwind CSS; back-end using Bun runtime + Express.js + TypeScript; database powered by PostgreSQL with Prisma ORM. Enabled real-time execution monitoring and analytics: workflows display live status updates, track success/fail rate, execution times, provide performance dashboard. Integrated multiple external services: Telegram Bot API, Resend API, Solana Transactions, OpenAI API and Webhook triggers for external system integration.

Built browser-based RAG app with multi-source ingestion (file upload: PDF/CSV/TXT; website URLs; and raw text), chunking, embedding, vector indexing, and semantic retrieval to power a chat UI with context-grounded answers. Implemented clean, responsive UI/UX with data source inputs, ingestion progress, RAG store status, and a streaming chat interface. Implemented LangChain pipelines, hybrid retrieval/reranking, Qdrant collection management, and SSE-based chat; documented APIs (/api/ingest, /api/chat, /api/user-files), provided Dockerized Qdrant setup, and outlined cloud storage strategy for serverless deployments.

Developed ZeroCV, an AI-powered interviewer platform with multi-role access (Recruiter, Candidate, Admin) using Next.js (Turborepo) and Node.js backend for a scalable monorepo architecture. Integrated Supabase authentication with Google login to provide secure, role-based access control and seamless onboarding. Built intelligent interviewer workflows with real-time voice agents and AI-driven assessments, enabling recruiters to evaluate candidates efficiently.

Engineered a full-stack AI-driven grading platform utilizing Next.js, MongoDB, Zustand, and Octokit for seamless GitHub API integration. Extracted and analyzed comprehensive GitHub contribution data, including commits, additions, deletions, pull requests, and active development days. Integrated large language models (LLMs) to assess code quality and automatically generate grades aligned with industry standards and best practices.