GenAI Engineer · Agentic AI · RAG Systems
Hey, I'm Priyanshu Patel — a software engineer building LLM agents, RAG and agentic RAG pipelines with LangChain and LangGraph, from retrieval design to API integration and deployment.
RAG and agentic RAG pipelines with embeddings, hybrid retrieval, reranking, grounding checks, and source citations.
Multi-step, tool-using agents in LangGraph: query routing, rewriting, self-evaluating retrieval, and parallel worker agents.
Tracing, prompt versioning, audit logging, and RAG evaluation so LLM behavior stays visible and debuggable.
FastAPI and Node.js services with Pydantic-validated structured outputs, async pipelines, and clean REST design.
Vector databases and hybrid search (vector + full-text, RRF fusion) feeding grounded, patient- or tenant-scoped retrieval.
Next.js and React interfaces plus Docker and AWS deployment, so GenAI features reach users end to end.
I'm a software engineer specializing in GenAI and agentic AI systems, with hands-on experience building LLM-powered agents, RAG and agentic RAG pipelines, and multi-step tool-using workflows with LangChain and LangGraph.
I build on a Python/FastAPI, Node.js/Next.js, PostgreSQL (pgvector) and AWS stack, and I'm comfortable owning a GenAI feature end to end — from retrieval and agent design to API integration and deployment.
"A good AI feature is grounded, traceable, and knows when to say 'I don't know'."
Orchestrator-worker LangGraph pipeline with an adaptive router (closed-book / hybrid / open-book), Pydantic-validated blog plans, parallel section writers via the Send API, citation-grounded generation, and on-demand Gemini diagram generation.
Agentic RAG that routes queries, rewrites follow-ups from conversation context, grades retrieved evidence, falls back to web search, and retries failed queries before declining. Answers are cited, and routing and evaluation decisions are logged for tracing.
Chrome extension for semantic Q&A over YouTube videos. FastAPI + LangChain RAG pipeline with Hugging Face embeddings, Qdrant cosine-similarity search, and Gemini, returning timestamp-cited answers. Backend deployed on Render.
Real-time AI voice agent using Sarvam STT/TTS with structured intent extraction. Low-latency conversational pipeline with WebSockets, caching, and a decision engine.