I'm a Forward Deployed AI Engineer who's slightly obsessed with making AI systems that actually work in production (you know, the hard part).

Currently at Supervity, where I get paid to argue with LLMs and occasionally win.

I build AI systems for enterprises and teach developers how to do the same. The best compliment I've received? "Your code actually works in production." (The bar is low, I know.)

Some numbers I'm proud of:

• Helped close $3M+ in enterprise deals (turns out, working demos > slide decks)
• Mentored 1000+ developers through Google Cloud community programs
• Shipped 50+ AI agents to production (and only mass-deprecated 3 of them)
• Cut idea-to-production time by 4x (my PM still doesn't believe me)

I got tired of explaining why a single chatbot can't replace an engineering team. So I built a system that actually tries to.

Project Agora is a multi-agent framework on Google ADK where specialized agents (analyst, researcher, architect, coder, reviewer) collaborate like a tiny autonomous dev shop. The code reviewer agent is ruthless — exactly like that one senior engineer we all know.

The interesting bits:

• Orchestrator treats agents as function calls in a DAG (determinism > vibes)
• BigQuery Vector Search for memory (because why add another database?)
• Human-in-the-loop checkpoints (AI writes code, humans approve it — for now)

Read the deep dive →

Ever had an interview where the silence felt eternal? That 3-second AI thinking pause? Yeah, I fixed that.

AIVA is a real-time AI interview coach with <500ms latency. It listens, thinks in parallel, and responds with lip-synced video — fast enough that you forget it's not human. Almost.

What made it work:

• Gemini 2.5 with native JSON mode (no more regex parsing prayer circles)
• Cloud Run Session Affinity (the WebSocket trick no one talks about)
• Safety Settings API (because an AI interviewer shouldn't roast candidates)

Read the architecture breakdown →

Instead of arguing about language performance on Twitter, I built a system where you can watch C++, Java, and Python race in real-time.

Spoiler: C++ wins. But Python has better snacks at the finish line.

What I learned:

• P99 latency doesn't lie (C++ at 5ms, Python at 25ms — sorry not sorry)
• Structured logging to Cloud Logging turns opinions into data
• Visualization teaches better than documentation ever will

Read the writeup →

Got tired of spending the first week of every project configuring boilerplate. So I built a template I actually want to use.

FastAPI + Next.js + PostgreSQL — all containerized with Docker. One command to start, zero environment debugging nightmares.

What's included:

• Python 3.11 backend with FastAPI, Pydantic, SQLAlchemy
• Next.js 15 frontend with React 19 and TypeScript
• PostgreSQL 15 with Alembic migrations pre-configured
• Docker Compose for consistent dev environments across any OS
• Code quality tools (Black, isort, ESLint, Prettier) ready to go

git clone https://github.com/MohitBhimrajka/template
cd template && cp .env.example .env
docker-compose up --build
# → Frontend at :3001, Backend at :8001/docs

View the template →

The Google Cloud stuff: Vertex AI, Cloud Run, BigQuery, ADK, Secret Manager, Cloud Logging (Yes, I drink the Kool-Aid. It's pretty good Kool-Aid.)

The AI/ML stuff: Gemini 2.5, LangChain, RAG pipelines, PyTorch, TensorFlow

The "I can also do this" stuff: Python, TypeScript, C++, Java, FastAPI, React, Docker, Postgres

I write about the stuff that doesn't fit in a README — architectural decisions, production war stories, and occasionally, opinions:

• Building an Autonomous Software Agency — Multi-agent systems on ADK
• Architecting a "Human" AI Interviewer — Latency, WebSockets, and controlled generation
• Polyglot Microservices on Cloud Run — When you need proof, not opinions

I believe you don't really understand something until you can teach it.

Lead, Code Vipassana (Google Cloud) — Helped 1000+ developers at Google's "Build and Blog" Marathon in Bangalore. Turns out, the best way to learn cloud architecture is to explain it to someone at 2 AM during a hackathon.

President, Technical Student Council — Ran hackathons and tech events at ATLAS SkillTech. Mostly involved convincing sponsors that students can, in fact, build cool things.

I'm always down to talk about:

• Why your RAG pipeline isn't working (it's probably the chunking)
• Multi-agent architectures (and why most of them are overkill)
• How to make AI demos that actually impress enterprise clients
• The eternal Python vs. compiled languages debate

Currently exploring: Forward Deployed AI roles where I can build and ship.

"The future of development isn't AI assistance — it's AI collaboration."

_{P.S. If you're hiring and made it this far, we should probably talk.}