Now
Updated August 2026. (This page is a work in progress.)
Work
Building LLM infrastructure at Oblivious. Current focus is on the multi-provider gateway — routing requests across Claude, GPT, local models, and other backends. Working on observability (tracking which model performed best for a task), cost optimization (fallback routing based on budget), and deployment tooling for confidential execution environments.
Also building deployment and attestation tooling for confidential computing — making it easier to deploy services to Nitro Enclaves and confidential GPUs with the same rigor we apply to regular infrastructure.
Learning
Deep diving into LLM ops — how to instrument models, interpret usage patterns, and build systems that make sense of what's actually happening at scale. Also exploring confidential computing more seriously; there's a lot to unpack about threat models, attestation, and how to actually reason about security in this space.
Reading more about distributed systems and consensus. Thinking about how infrastructure lessons apply to the LLM layer — resilience, fallbacks, observability, cost optimization.
Notes
This page is inspired by the nownownow.com convention. It's meant to be brief and genuine, not exhaustive. I update it occasionally when my focus shifts meaningfully.