Reconstructing poker hands from plain English + solver-backed analysis

Hi everyone,

I've been working on a side project that tries to bridge the gap between solver output and how people actually talk about hands.

I think there's a big gap between GTO numbers and memorizing solver output versus how people actually learn and review hands. So I built a pipeline where I feed solved poker nodes to an AI model that's been trained to interpret the game context and infer meaning from those numbers. rather than just restating them. Where solver coverage exists, it’s grounded in real solves; where it doesn’t (weird stacks / configs), it will try to reason from similar structures.

The idea is to have something you can go to quickly in-game or right after playing a tough hand. Not to get a super specific "Bet 52.5% here" type chart, but more to quickly understand "Was that a punt? Why does solver like to check more often in this spot?"

Anyways, I got the pipeline to work pretty well so far and got some interesting results, but I need someone to help me test it and give me feedback on how I can improve it further.

For example, you can paste something like:

Pretty much like a normal post or how you'd talk in discord chat. It'll rebuild the hand, sync it to a table view, and explain why solver prefers certain lines and show you the stats etc... e.g. which hands we're actually value betting, which hands beat us, and where mistakes tend to creep in.

I'm mostly looking for people willing to give it a try and stress-test it with real, ugly spots and tell me where the reasoning is off, what else could make it more convenientnt to use for study. I'm especially interested in edge cases (paired boards, weird SPRs, river decisions, etc.).

If anyone wants to throw a hand at it, it's here:
👉 https://railbird.gg

Appreciate any feedback, good or bad.