Back to the Keyboard

This is my GitHub contribution graph for the last year.

You can see exactly when life took over.

September to November is solid green. I was shipping most days, usually on ModelFitAI. Through the winter it thins out, and by April it's patchy. May to August is almost entirely grey.

There's one square in May. That's the day I published my last post on this blog.

I Missed It

Nothing dramatic happened. Work got busier, home needed more of my time, and like most people juggling both, the side project was the first thing to slide off the end of the week.

I won't dress it up. I stopped building, and I missed it more than I expected to.

Where ModelFitAI Is Right Now

To be clear about what I'm re-pivoting from: ModelFitAI was already an agentic platform. People were deploying agents on it and using them: lead gen bots, autopilot posting, launch-day trackers, all running from Telegram. That part worked.

The two lonely squares at the end of the graph are both Sundays. Both times I was taking things out rather than adding them.

ModelFitAI started as a model recommendation engine: describe what you're building, and it tells you which LLM fits best. In August I deleted it. It was the founding idea and it's in the name, but it wasn't what people came for. They wanted the agents.

And the agents themselves had a dependency I never said out loud to customers. They ran on a third-party agent framework I built on top of rather than built myself. In September I stripped that name off everything customer-facing. If you use ModelFitAI today, you won't see it referenced anywhere. It's still doing work underneath, the way a checkout page can run on infrastructure it never names out loud. But the product, the story, and what I owe an explanation for are mine now, not a framework's.

Paid plans and the free tier are also paused while I do this properly. Right now ModelFitAI is effectively on hold, and that's deliberate.

The Re-Pivot

I'm re-pivoting ModelFitAI again. Same idea, an agentic platform your ICP can actually run agents on, but this time it's built on my own footing instead of someone else's name, and shaped by what the people already using it actually needed, not by what I assumed they'd need.

I have plenty of ideas about what that looks like next. But I've already built one thing nobody asked for (the recommender), and I'm not doing that twice.

Starting With 91 People

There are 91 people on the ModelFitAI waitlist. They signed up, some of them months ago, and then heard nothing from me while the squares went grey. That's on me.

So that's where I'm starting. Before I write any new code, I'm going to write to every one of them. I'll ask what they signed up hoping for, and what would make them actually use it.

What they tell me will shape what gets built.

If You're on the Waitlist

Look out for a message from me this week. It's the first ModelFitAI update, and you'll get it before it's posted anywhere else.

Please reply to it. That's the whole point.

If you're not on the waitlist and want to be part of this, you can join on the ModelFitAI home page.

Slowly, and in Public

I'm not promising daily commits. I tried that pace once, and the graph shows how it ended. Updates will go to the waitlist first, then here.

And if you're looking at your own grey squares right now: they aren't a judgement on you, just a record. Mine is going to look different by next September.

Pravin Boppuri is a Senior Architect at Henkel AG with 18 years of enterprise technology experience across manufacturing, supply chain, and financial services. Outside his day job he builds AI systems and writes about the architecture decisions behind them. He is the founder of ModelFitAI, a no-code agent deployment platform built on the Claude API.