- Tools
- Claude Code, Figma, spreadsheets
- Time
- 1 week
- Status
- Personal tool, runs locally
- Guardrails
- All range data from me as JSON, plus a design system, CLAUDE.md file and an overall project plan



Context
I wanted solver-grade poker training for my own use that wasn’t locked behind a paywall.
What I delegated, and the calls I made
Marked rows are key decisions.
- Writing the code
Claude Code wrote it from my plan and mocks
Mostly AI
- Sourcing the range data
All of it came from me
All me
- Cleaning the data
Claude helped clean it up
Shared
- App design
My Figma mocks guided the build
Mostly me
- Table positions Override
Claude wanted 6 seats; I required up to 9
All me
- Tradeoff:
- more ranges to source and more layouts to design.
- Why:
- training should mirror the games I actually play.
- Scope Decision
Preflop only for v1
All me
- Tradeoff:
- the trainer covers just the start of a hand.
- Why:
- postflop scenarios multiply the data far beyond what I wanted to take on for a first version.
Outcome
It runs fast and works well on my laptop. Next, I’d consider adding postflop training.
