A compact map of what I build and what I spend time thinking about.

I'm Rakesh Sawant — an AI/ML engineer who builds retrieval systems, GenAI pipelines, and enterprise intelligence that holds up in production.

Most AI systems are impressive in demos. Fewer survive messy data, shifting requirements, and the complexity of real software ecosystems. That gap is where I work.

The work spans synthetic data, Hybrid RAG, knowledge graphs, and semantic retrieval — designed to be reliable, not just functional.

I'm currently an AI/ML Engineer at Strategy (formerly MicroStrategy), embedding AI capabilities into enterprise business intelligence platforms.

The work sits across retrieval systems, model orchestration, scalable AI services, distributed data workflows, and backend APIs — making ML features behave reliably inside large, complex software ecosystems.

Outside work: system design rabbit holes, Graph-RAG experiments, agent architectures, and a persistent interest in making AI systems more observable than mysterious.

• 2026-07-18 9 min read
  Multi-Agent Orchestration: Choreography, Circuit Breakers, and the Coordination Problem Nobody Designed For Wiring three LLM agents together isn't an AI problem, it's a distributed systems problem. Orchestration vs. choreography, agent-level circuit breakers, and why shared state between agents will bite you exactly the way it bit microservices a decade ago.
• 2026-07-05 18 min read
  The Stateful Agent Problem: Why LLM Agents Are Just Distributed Systems in Disguise Multi-step LLM agents aren't an AI problem. They're a distributed systems problem wearing an AI costume. State consistency, idempotency, saga patterns, checkpoint-and-replay: every hard lesson from the microservices era is showing up again, and most agent frameworks are pretending it isn't.
• 2026-06-18 20 min read
  The Token Budget: Decoupling Rate Limits from Request Counts Your AI feature has 10,000 users and a 90K TPM ceiling from OpenAI. One power-user can bankrupt your entire API budget in under a minute. Here's how to build token-aware rate limiting that actually holds up.
• 2026-06-14 12 min read
  The Async AI Processing Pipeline: Decoupling Storage from Intelligence Why treating ML pipelines like HTTP requests will destroy your production system, and how to build AI infrastructure that actually scales.