Applied AI engineer. I build agent systems that run unattended, heal themselves, and cost close to nothing to keep running.
New Delhi, India · Open to AI/agent engineering roles
Most LLM systems work in a demo and quietly rot in production — a page changes, a prompt drifts, a bill triples, and nobody finds out for two weeks. I build the opposite: agents that detect their own breakage, repair themselves, and spend tokens only where a model is genuinely needed.
🕸️ Self-healing scraping agent
An agent (headless Chrome via CDP + an LLM) that writes its own extraction recipe once per site, then replays it with zero model calls. When a site changes and the recipe stops matching, it treats the empty result as a failure signal, not "no data", and re-authors itself automatically.
agents · CDP · self-healing · cost routing
✅ Multi-agent to-do platform
Task management as active execution, not passive tracking. Tasks are dynamically decomposed, delegated, scheduled, and partially or fully executed by specialized autonomous agents.
multi-agent · orchestration · planning
🎬 LLM-to-video pipeline
Prompt → parametric SVG → rendered video. Every stage is schema-validated (Zod) with an automatic repair loop, so a malformed model output gets fixed instead of crashing the render.
structured output · Remotion · self-repair
| Languages | TypeScript Python JavaScript SQL |
| AI / ML | LLM agent design Gemini tool use ONNX Runtime INT8 quantization transformers OpenCV |
| Backend | NestJS Prisma PostgreSQL Socket.io REST WebSockets |
| Infra | Docker Chrome DevTools Protocol Playwright Linux Git |
Finishing my degree while working full-time on agent infrastructure.
I'm looking for: a small, high-ownership team building agent/LLM systems that ship to real users — where I own a system end to end rather than a slice of it.
If you're building agents, or you have an LLM system that works in a demo and breaks in production — I'd like to hear about it.
