Hari chandra prasad

I’m a Backend Engineer Intern focused on shipping production-ready AI/LLM capabilities—especially RAG and agentic workflows. At Metry AI, I built an LLM agent using OpenAI function-calling that autonomously selects among 5 read-only database tools and chains multi-step queries.

I also engineered a secure read-only data layer with parameterized Supabase queries and server-side per-user scoping, eliminating SQL injection risk and preventing cross-user data leakage. To improve performance, I implemented Upstash (Redis) caching that reduced response latency by 70%+ on high-frequency API routes, and I added backend logging/observability to shorten debugging during production issues.

Earlier, as an AI/ML Intern at BRC Web, I delivered an end-to-end pet breed recognition pipeline: a TensorFlow/Keras CNN, a Flask REST API for inference, and real-time confidence scoring with a low-confidence fallback integrated into a React frontend. In my projects, I’ve built a RAG Documentation Assistant (web crawl → chunking → Pinecone → conversational retrieval with source citations) and multiple LangGraph-based agent systems with self-correction and reflection loops.