Prathamesh Mane

At Pelocal, I build production conversational AI systems for transit and emergency dispatch workflows. For a Delhi Metro client, I'm building a LangGraph-based assistant for route planning, fare lookup, and station queries serving 10,000+ monthly users. I architected state graphs, checkpointing, tool-calling, RAG, memory, and evaluation workflows for reliable multi-step interactions. I also built a real-time roadside-assistance voicebot from scratch, connecting streaming speech recognition, LLM routing, multi-provider TTS, and dispatch APIs. The system delivers sub-one-second end-to-end voice latency and reduced service fulfillment time from eight minutes to two minutes across 500+ daily production calls. I work across Python backend and AI infrastructure, including FastAPI, Redis, RabbitMQ, WebSockets, LangGraph, OpenAI SDK, and model evaluation. I'm especially interested in reliable agentic systems that perform under real-world constraints.