Infragen
The control plane for AI infrastructure: deploy, monitor and operate agents and models from one visual console.
Infragen gave teams and individuals one place to deploy, monitor and manage AI infrastructure: a visual, node-graph console for wiring up and operating agents and models. I led the frontend and built deep into a typed Python/FastAPI backend.
At its core was a multi-agent systemI led: Pydantic AI and Smolagents orchestrating multi-step jobs reliably, with structured error recovery so long-running work didn't fall over.
- Role
- Full-stack · frontend lead
- Duration
- 2024 — 2025
- Surface
- Console + marketing site
- Stack
- Next.jsReact QueryReactflowPropelAuthPythonFastAPIPydantic AISmolagents
The hard parts.
A Reactflow canvas to wire, deploy and monitor agents and models: drag-connected flows with live status, backed by real-time state.
Led a Pydantic AI + Smolagents system running multi-step jobs:92% accuracy across 1,000+ monthly jobs, with task time cut ~40% through fault-tolerant agent communication.
PropelAuth for organisations and individuals, gating a console that operates live infrastructure, with state kept fresh via React Query.
A FastAPI + Pydantic backend with end-to-end typed contracts, so the console and the services behind it never drifted out of sync.