Federico Kamelhar

Building AI that works in production.

I build AI agents and the platforms they run on. Here I write about agent architecture, orchestration, and infrastructure — what works, what breaks, and what I learn along the way.

Writing 2026
  1. 30 Sep

    An agent kill switch you can run

    NVIDIA's OpenShell, an agent doing real work on OCI Generative AI and OpenRouter, and three ways to stop it without asking it — measured.

    safety · 13 min read

  2. 20 Sep

    Building local inference on a DGX Spark

    A 35B model under k3s, one OpenAI-compatible endpoint in front of it, and the eight steps in between — with what each one taught me.

    infrastructure · 30 min read

Work
  1. Tulip Agents open-source agentic framework · building tulipagents.ai
  2. GSAR typed grounding for hallucination detection and recovery arXiv
  3. Open-source contributions OCI support in LiteLLM, NeMo and vLLM production-stack github
  4. Strands Agents led its creation at AWS · open source github