How I Built One Control Room for Six AI Agents
What changed when I separated personal, work and general-purpose AI agents, then made their health, token use, cost and recent activity visible in one private Grafana control room.
Rajeev Gill
I build practical software around AI, data, analytics, and adtech.
My work sits between product thinking and implementation, with a bias toward systems that make real work less messy.
Based in London. Interested in useful AI, clearer measurement, and software that holds up outside a demo.
Get in touchSelected projects
Things I have built
Research product
Open GTM Index
A transparent guide to open-source sales and marketing software.
TanStack StartReact 19TypeScript
AI research product
Local LLM Lab
A measured guide to every local AI model installed on my Apple Silicon workstation.
PythonReactVite
Operations product
Hackathon Voting App
A production-ready single-screen judging app built for a live hackathon room.
Next.js 14TypeScriptClerk
Recent writing
Notes from the work
I Made GLM-5.3-Flash 63% Faster. I'm Keeping the Cheap Route.
A small OpenRouter routing benchmark made long GLM-5.3-Flash generations 63% faster, but modelling the result against a real day of agent use made the cost trade-off much clearer.
My AI Agents Spent More on Oversight Than on Work. Here's the Data That Made Me Delete the Middlemen.
I ran four layers of orchestration between a request and actual work. The telemetry said most of it was overhead. Here is the before, the after, and the numbers that convinced me.
Have something useful to build?
I am always interested in thoughtful software, analytics, and AI work where the practical details matter.
Email me