AI-Automation & Systems Engineer
2025 – PresentVisions, Chandigarh · InsureTech client (UK)
- Architected and built a two-phase automated product-validation pipeline that shifts insurance-rate validation left from engineering to product managers, catching data errors before they reach development.
- Designed a server-side LLM first-pass reviewer (headless Claude) running a 5-checkpoint review over raw insurer files with image- and checkbox-aware PDF parsing — generating a plain-language report and executable test cases at ~$0.75 and ~5 minutes per product.
- Built the developer-stage automation: generated Playwright tests from raw rate/benefit files, ran them against the live quoting engine, and versioned certified test cases in S3 as the QA contract.
- Quantified impact via a retroactive pilot across 7 production tickets — up to ~58% of historical dev↔PM clarification cycles automatically catchable — with bit-exact rate derivation validated on real insurer data and zero false positives across 4 shipped bundles after calibration.
- Hardened the pipeline for horizontal scale: shared BullMQ/Redis queue (exactly-once execution across a load-balanced fleet), per-product idempotency, and surgical cross-instance job cancellation — verified live with zero duplicate or orphaned processes.
- Conceived and ran an automated Figma → Locofy → GitHub → AI → Pull Request workflow: 51 automated design-to-PR runs across 3 modules in ~6 weeks, ~8 minutes per page, preserving business logic and coding standards.
- Evolved that pipeline into a closed-loop, self-verifying design-to-code agent: a single `/locofy <ticket>` command reads a Jira ticket (Atlassian MCP), pulls the live Figma via the Locofy MCP, implements into React/TypeScript under an 'architect' rule, then self-verifies in a rendered Playwright loop (visual + exact-token diff) before opening a PR — validated end-to-end, targeting ~2–3 dev-hours/page vs ~12–16 hand-coded.