Janakiraman K

Most AI engineers build wrappers. I build the engine underneath.I'm an Agentic AI Engineer who spent the last 2+ years making LLMs actually useful — not just responsive. That means systems that plan across steps, delegate to tools, retain context across sessions, and ship to production without falling apart.Here's what that looks like in practice:→ Built production RAG pipelines achieving 90–93% retrieval accuracy — 30%+ above keyword-only baselines — using FAISS, hybrid retrieval, and semantic chunking→ Cut LLM inference latency by 20–30% through async agent execution and prompt routing on GCP Vertex AI + GKE→ Designed multi-agent orchestration workflows with LangGraph and AutoGen that handle real-world complexity: branching decisions, tool delegation, and stateful memory across long sessions→ Won 11 hackathons (5x Web3, 6x Web2) — competing and shipping under pressure, not just prototypingI don't just study agentic AI. I debug it at 2am when it behaves non-deterministically, ship it when it's ready, and iterate when it's not.What I work with:Agentic frameworks: LangGraph · AutoGen · LangChainInfrastructure: GCP (Vertex AI, GKE) · Docker · FastAPIRetrieval & memory: FAISS · pgvector · semantic chunking · hybrid searchLanguages: Python · SolidityCurrently building: AutoAgent (NL-to-agent executor) · G3 Dev Hub (decentralized GitHub alternative)Currently open to full-time AI/ML Engineering roles — particularly teams working on frontier agent systems, LLM infrastructure, or production-scale AI products.If you're building something that needs more than a wrapper, let's talk.Reach me at janakiramankeerthivelan1021@gmail.com or connect here on LinkedIn.