I build AI systems that actually work in production — retrieval pipelines, agentic reasoning loops, real-time voice/WhatsApp interfaces, and the full-stack layers that connect them.
Currently open to AI Engineer and Full Stack roles at early-stage startups and remote-first teams.
MemoraAI — Hybrid RAG with pgvector + BM25 + Reciprocal Rank Fusion + cross-encoder reranking. Three-tier memory architecture. Built on Next.js, FastAPI, and Supabase.
TraceAI — A ReAct agent built from scratch, no LangChain, no abstractions. Manual reasoning loop, tool execution, and full step-by-step tracing. The point was to own the logic, not wrap it.
SpeakKrishna.ai — Conversational AI over the Bhagavad Gita using RAG over verse data. Built solo over a weekend with Kimi/Gemini + Supabase/pgvector.
| Layer | Tools |
|---|---|
| AI / Retrieval | RAG, pgvector, BM25, RRF, cross-encoder reranking, ReAct agents, OpenAI, Gemini, Mistral, Whisper |
| Backend | FastAPI, Node.js, NestJS, Express.js |
| Frontend | Next.js, React, TypeScript, Tailwind CSS |
| Databases | PostgreSQL, pgvector, MongoDB, Supabase |
| Infra / Other | WebSockets, Plivo, Sarvam AI, Docker |
- Building systems I understand end-to-end, not just gluing APIs together
- Retrieval quality over retrieval speed (though ideally both)
- Async, remote-first, ownership-driven teams
📫 linkedin.com/in/ganesh-meti
🐦 @Ganesh_A_meti
🌐 Portfolio
Always building.