📍 Bengaluru, India | 🤖 Director of AI Engineering @ Certa | ⚡ Ex-Frontend Engineer turned AI builder
8+ years building software and teams. Started deep in the frontend trenches — React architecture, TypeScript migrations, test coverage wars. Now I lead AI product and engineering teams, figuring out how to make enterprise software genuinely intelligent. I care about shipping real products with LLMs, not just demos. And yes, I have opinions about whether Claude actually follows your CLAUDE.md.
Director of AI Engineering at Certa — a third-party lifecycle management platform. I've been here 7+ years, which in startup years is basically a lifetime.
I own the AI product strategy and engineering org. That means deciding what we build, how we build it, and making sure the teams have what they need to execute. Concretely:
🧭 AI Product Direction — Defined and shipped Certa's entire AI product lineup from zero: NLP-powered workflows, automated document analysis, intelligent process automation. These aren't experiments — they're in production, handling real enterprise workloads.
📈 Org Building — Scaled the engineering org from 10 to 70+. Hiring, org design, establishing engineering culture, performance frameworks — the full stack of people work.
🏗️ Technical Foundation — Earlier in my tenure, I was hands-on driving the technical bets that paid off: TypeScript migration, monorepo adoption, TDD culture (took test coverage from 15% to 85%), and a 36% LCP improvement across all pages. I still code, but these days it's more about setting technical direction than writing PRs.
Before Certa: Level AI (AI-powered customer service intelligence — QA automation, real-time agent assist, conversation analytics) and YourMech.in (CTO & Co-founder).
- 🧠 mind – Save articles. Read them when you're ready.
- 🛡️ auto-guardian – CLI to generate test cases for React components using AI
- 💬 chat-notes – Transform scattered thoughts into organized intelligence
- ✍️ logs-and-gains – Published version of my writings
- 🔄 lebab-as-babel-plugins – Codemod to transform ES5 to ES6/7 using babel plugins
On agent instruction files (CLAUDE.md, .cursorrules, etc.) — I'm not fully sold. The core issue is non-determinism: these files can be ignored or only partially followed by the model, and the misses are easy for humans to overlook unless you're explicitly auditing every output. We're building workflows on top of a foundation that doesn't guarantee compliance with its own instructions. That tension is underexplored.
On AI-assisted development — I'm deep in the Claude Code / agentic coding world. I track memory usage patterns, test whether models actually follow TDD instructions, and share findings when things break. I'm interested in the gap between what these tools promise and what they reliably deliver in production codebases. I repost people like Simon Willison and follow the "Software Factory" discourse closely — the idea that code shouldn't be written or reviewed by humans is provocative and worth stress-testing.
On React at scale — After years of running a large frontend org, I think about monorepo architecture, design systems as engineering leverage, and where the React ecosystem is headed. I took the React Compiler into production code and spoke about what I found at React India 2024.
On LLMs in enterprise — Most LLM demos are impressive. Most LLM products in production are fragile. I think about the gap between the two constantly — how to build AI features that are reliable, auditable, and actually useful to non-technical end users.
✍️ Blog: rajatvijay.in — writing about engineering, AI, and everything in between.
🎤 React India 2024 — "Experiments with React Compiler in Production Code"
🎤 React Nexus 2024 — "Experiments with AI-Generated Tests"
Building at the intersection of frontend craft and AI ambition. Currently vibing with Claude Code and questioning everything about agentic workflows.