Full-stack engineer at ContextQA, strongest on the backend. I build the core QA-automation product's backend — test-execution engine, VNC streaming, multi-cloud browser orchestration (Playwright / WebdriverIO / LambdaTest) — and I ship the UI too: both products I shipped there are React / Next.js / TypeScript, one of them an accessibility auditor.
What I care about: idempotency, concurrency control, and testing that actually proves something.
Portfolio: shaileshchaudhari.vercel.app · How I verify: /engineering — real defects from my own production code, each with why the wrong version looked correct.
| Project | What it is |
|---|---|
| BALLAST | Deterministic simulation of a multi-tenant session control plane — per-tenant concurrency caps, quota windows, leases over an unreliable substrate. One integer seed in, one byte-identical decision log out. Judged by 8 invariants, a reference oracle, and mutation testing; it found real bugs, about half of them in the checker rather than the implementation. Zero runtime dependencies. |
| EduScale | Real-time coding battles where the two players' events may land on different Node instances — Socket.io Redis adapter, Redlock battle-start mutex, circuit breakers, Prometheus metrics. Live. |
| KhataGO | WhatsApp-first bookkeeping. Meta delivers webhooks at least once, so idempotency is enforced in Postgres — a unique message id plus an atomic PENDING→PROCESSING claim — with a containerized polling worker and a dead-letter queue with audited replay. |
| promptproof | Zero-dependency LLM eval + regression-diff kit — define cases, grade outputs, save a baseline, fail CI on any pass→fail. |
| idempotency-kit | The Stripe Idempotency-Key pattern as a zero-dependency wrapper, including the fingerprint-mismatch guard, plus a sliding-window rate limiter. |
| grounded | Production-grade RAG starter: cited answers, idempotent ingestion, an "I don't know" guardrail, and an eval harness that runs offline. |