Bharat Khanna's Blog

A high-level framework for integrating agentic AI with real-time analytics pipelines, covering architecture, governance, trade-offs, risks, and evaluation across finance, IoT, and healthcare.

Not all problems an AI agent solves are the same. This piece unpacks the distinction between Skills (knowledge layer) and MCP (execution layer) — and why separating them leads to more efficient, maintainable agent systems.

Everything you need to go from a working LangGraph agent to a production deployment: containerization, async serving with FastAPI, observability, circuit breakers, and zero-downtime deploys.

A production-grade guide to agent evaluation: single-turn unit evals, trajectory scoring, multi-turn simulation, harness engineering, and LangSmith-driven continuous improvement loops.