Suvom Karak

Built and productionized an internal-assistant platform using an agentic FastAPI gateway, including session identity, graph lifecycle, state persistence, and streaming. Designed LLM orchestration with LangGraph using structured outputs, follow-up-aware routing, tool-based handoff, and SQL query validation with self-correction. Implemented human-in-the-loop control with LangGraph interrupt/resume and durable execution via Postgres checkpointing. Added request guardrails, MCP tool discovery, and production observability with Langfuse and OpenTelemetry tracing plus model-quality evaluation.

Developed production LLM applications on Azure OpenAI for enterprise document processing and semantic search across large knowledge bases. Engineered RAG and semantic-search pipelines using vector similarity indexing and built automated extraction, chunking, and categorization. Delivered backend services for document analysis and multilingual translation/summarization using Flask/FastAPI, integrating with internal systems under security constraints. Created an end-to-end agentic pipeline for recruitment operations with LangChain and supported model evaluation, deployment, and ongoing production monitoring of quality and latency.