Ramnath T

Software Engineer

Siemens • Full time

Aug 2024 - Present • 2 yrs 1 mo

I build AI agent systems and automation for a living: multi-agent architectures, MCP tooling, and the evaluation harnesses that keep them honest. Two years building enterprise GenAI at a global engineering firm. I architected a 35-agent, 21-skill multi-agent framework automating code review, guideline enforcement and requirement engineering across the development lifecycle, and designed a self-upgrading AI knowledge base with drift detection, citation provenance and a 17-test deterministic evaluation harness built to cut stale context and hallucinated output. Earlier I built a source-code knowledge graph in Python and SQLite, indexing 1,884 files into roughly 30K nodes and 37K edges, exposed to agents through 9 MCP tools. Open source: merged fixes in qdrant/fastembed (the embeddings library behind many vector-search stacks) and opensquilla. Open PRs at mcp-atlassian and newspaper4k. What I build for clients: - AI agent workflows: LLM routing and classification, human approval gates, multi-step orchestration - n8n and Python automation: API integration across CRMs, messaging, sheets and internal tools - MCP servers, so your agents can actually reach your systems - Evaluation harnesses, so you can tell whether the AI is working rather than just running I write the test before I claim something works. On two open-source bugs I stopped at "reproduced" and went back to the maintainer rather than ship a fix that would have silently corrupted data. In one case the maintainer independently confirmed the call.