I build tools for people and models that need to be right, not just plausible.

Most of my current work sits around grounded AI tooling, Rust-heavy developer infrastructure, and runtimes that keep correctness in the loop instead of pretending the model can guess everything in one pass.

yoyo is my main current project: a local MCP server and Rust code-intelligence tool for grounded repository reads, ownership judgment, and safer code changes.

Right now I am pushing it toward a stronger product shape: better evals, better terminal UX, and better support for recursive subproblem decomposition instead of shallow repo-tool glue.

This is a transformer runtime lab built around problem-shaped virtual machines.

The thesis is straightforward: code should own legality, state transitions, and rollback; the model should score ambiguous branches on top of that runtime, not replace it with a one-shot hallucination.

I have been actively contributing around Helm install behavior, deployment defaults, chart docs, and health-check reliability.

I like this kind of systems work because it forces the docs, templates, and actual runtime behavior to line up.

• fluid-simulator-webapp: a browser-based 3D particle simulator built with TypeScript, Vite, and Three.js.
• OSS contributions in projects like Hatchet and Iggy.

• Grounded AI tooling
• Rust developer tools
• Runtime and systems design
• Cloud and platform infrastructure
• Products with strong correctness boundaries

• GitHub: @avirajkhare00
• X: @avirajkhare00

If you are building serious AI infrastructure, developer tooling, or systems software, I am probably interested.