Production LLM observability layer — Langfuse tracing, p50/p95/p99 latency, cost-per-request, LLM-as-judge quality scoring (4.4/5), and CI benchmark gate on every PR.

Open-source prompt injection firewall + hallucination blocker + agent memory layer for any LLM app

RAG-based document Q&A system — ask questions from any PDF using LangChain, FAISS, and Groq LLM

RAG-based policy assistant for bank collections compliance — hybrid retrieval (dense + BM25), cross-encoder reranking, and layered guardrails with groundedness checking. Runs locally via Ollama, zero cost.

RAG-powered customer support chatbot with intent detection, multi-turn memory, and Groq LLM