Now

Building

Actively improving PR Scrutiny — specifically making the non-AI static analysis layer more robust. The LLM agent is the easy part; the hard part is building deterministic, fast analysis that catches real issues without false positives. Researching better approaches to import graph traversal, entropy-based secret detection, and complexity heuristics.

Learning

Going deep on how LLMs actually work — not the math, but the systems and design decisions. How attention patterns form, why certain prompting strategies work, what actually happens at inference time. Understanding the tool well enough to build on top of it without guessing.

Alongside that: large-scale system design. How systems stay fast and cheap under load — caching strategies, database tradeoffs, the cost of every extra network hop. This directly feeds the goal: more users, lower costs, better experience.

Looking for

Full-time role — remote. Building products that real people use, on a team that cares about the craft. AI infra, developer tooling, or any product-focused engineering role where I can talk to users and ship things that matter to them.

If something I've built is relevant to what you're working on, reach out.