Hi, I'm Harsh.
A freelance AI engineer who takes products from napkin concept to production, in days, not quarters.
I'm a freelance AI engineer and systems architect based in Bengaluru, India. I partner with founders, high-growth startups and visionary teams around the world to take AI products from napkin concept to production reality.
My foundation was forged at enterprise scale: architecting AI agent pipelines at Allianz Technology, leading generative AI R&D for engineering automation at Dassault Systèmes, and building IoT software at UST. Today my focus is rapid zero-to-one execution, combining LLM reasoning, vector memory layers and smooth React 19 interfaces.
I bridge the gap between cutting-edge LLM research and production software
Reliable autonomous agents, low-latency streaming interfaces and resilient data pipelines that founders can scale.
Fact sheet
- Role
- Freelance AI engineer and systems architect
- Based
- Bengaluru, India
- Currently
- Senior AI Agent and Software Engineer at Allianz Technology
- Before that
- Dassault Systèmes and UST
- Education
- B.Tech, Computer Science, ITER Bhubaneswar
- Availability
- Select contracts and advisory, replies within 12 hours
Kumar Harsh is a freelance AI engineer and systems architect based in Bengaluru, India. He has built AI agent workflows at Allianz Technology, led generative AI R&D at Dassault Systèmes, and shipped ten AI projects in the last six months.
Start a projectCareer timeline
From campus to production AIITER, Bhubaneswar
Graduated with a focus on computer science, mastering the fundamentals of data structures, algorithms and distributed web systems.
UST
Developed complex web portals for IoT-based connected vehicle projects. Pioneered the internal ReactJS starter kit used across multiple teams.
Dassault Systèmes
Led research into generative AI for industrial design. Built autonomous agents that navigate complex engineering schemas and automate R&D workflows.
Allianz Technology
Architecting resilient AI agent workflows, enterprise LLM orchestration and automated ML pipelines for core global financial and insurance operations.
How I work
Production, not prompt scripts.
Full-stack AI with streaming UX and vector infrastructure, not just notebooks.
Keep a human in the loop.
Review guardrails, token budgets and observability on every agent I build.
Plan for provider failure.
Fallback cascades, retry budgets and circuit breakers, so a rate limit never becomes an outage.
Ship in weeks, not quarters.
A zero-to-one sprint takes 10–14 days, with CI/CD ready on day 14.