agents that don't hallucinate on my watch
AI Systems Engineer — I take ambiguous, undocumented problems and ship production AI systems: agentic platforms, retrieval infrastructure, and the eval/observability layers that make them trustworthy.
I care less about demos and more about what breaks at 2am — and how fast I can find out why.
- 0 → 1 AI systems: stakeholder discovery → architecture → build → deploy → eval → iterate, owned end-to-end
- Agentic infrastructure: LangGraph multi-agent systems, MCP servers, tool-calling harnesses, retrieval pipelines (RAG, Text-to-SQL, hybrid search)
- Production reliability: distributed systems on Kubernetes, GitOps, observability (Prometheus/Grafana/OTel) — the boring infra that makes AI systems trustworthy, not just demoable
- Evaluation & trust: deterministic agent testing, regression harnesses, precision/recall/F1 pipelines — I don't ship what I can't measure
aegis-harness — a deterministic eval + regression testing engine for LLM agents (because 95% of agent projects fail from non-deterministic behavior nobody caught in CI)
Deeper agent-orchestration patterns at scale — multi-region deployment, cost-aware model routing beyond single-cluster setups
Open-source agent tooling, eval frameworks, or MCP ecosystem projects — if you're building developer tools for the agent era, I want to talk
| Project | What it does |
|---|---|
aegis-harness |
Deterministic evaluation & regression testing for LLM agents — trajectory assertions, tool sandbox, pytest + CLI integration |
nexus-mcp |
Multi-tenant MCP gateway with hybrid retrieval for enterprise document intelligence |
📫 Let's talk: LinkedIn · jassay2711@gmail.com · Pune, India
