I'm Rishi — a senior engineer with thirteen years of shipping production software, most recently as tech lead of the web team at an AI search-intelligence company, where I owned an extraction pipeline handling millions of events a day, built the single sign-on end to end, and led the design-system migration. I also took AI code review from a one-week pilot to the standard across the whole engineering organisation, and built the Claude Code tooling my team used for incident triage and release work.

In 2026 I started Tankster AI to work on the problem I kept running into: an autonomous agent is only as trustworthy as the structure around it. So I build that structure — a fail-closed egress proxy in Rust that redacts and meters everything an agent sends to a model vendor, a policy gateway for MCP tool calls, and an adversarial harness that judges an agent by what actually changed on disk rather than by what it said. All of it has tests, all of it has CI, and all of it says plainly what it does not do.

I care about the unglamorous things that separate a demo from something people rely on: verification that comes from outside the thing being verified, budgets enforced at a gateway rather than in a prompt, and documentation honest enough to survive a sceptical reader.