I'm an AI Full-Stack Engineer building production-grade AI systems — RAG pipelines, multi-agent architectures, and the infrastructure that actually keeps them running.
Currently pursuing MCA in Data Science at Parul University while founding VDev Automations, an AI automation agency. I don't just prototype — I deploy, evaluate, and debug until the numbers are honest.
What I've shipped:
ContextQuery — a hybrid RAG system (BM25 + semantic search + Reciprocal Rank Fusion) hitting 100% retrieval precision and 87.5% answer faithfulness, deployed on free-tier infrastructure using NVIDIA NIM, Chroma Cloud, FastAPI, and Next.js 15.
CyberRescue — a locally-hosted MCP server that lets Claude debug live Docker containers. Security-hardened with command validation, concurrency caps, and 16 passing unit tests. Listed on Glama and awesome-mcp-servers.
AI Coding Mentor — in progress. Multi-agent LangGraph system (Router → Analysis → Mentor → Execution → Evaluation) with Docker sandboxing, LangFuse observability, and full-stack deployment.
What I write about:
The decisions nobody documents — why I chose RRF over naive retrieval, what broke in production that didn't break in dev, and how to evaluate AI systems with actual metrics instead of vibes.
If something I built solved a problem you're hitting, the writeup is here. If you want to collaborate or hire someone who's already shipped production AI — reach out.
Certifications: Anthropic (Claude API, Claude 101) · Google (Gemini/Imagen, Vertex AI Prompt Design)
GitHub: github.com/vivekpatil200320