Static chart viewers show you frequencies without context. AI Poker Lab combines an O(1) pre-computed spotHash database with a native multi-threaded Rust 6-Max & Heads-Up Linear MCCFR blueprint, a PyTorch ReBeL belief-state value network, Burch safety-gadget continual resolving, and an IR spaced-repetition drill engine that closes your EV leaks.
24 free 6-max preflop charts · 18/18 validated GTO benchmarks · Post-hand study only
Enter your email to get instant access + weekly strategy tips.
O(1) spotHash · 15.96M+ Infoset Blueprint
Built from scratch in Rust, PyTorch, NumPy, and TypeScript—every component is benchmarked with exact turn/river best-response probes, AIVAT control variates, and offline Information Retrieval metrics.
Parallel 6-Max & Heads-Up Linear MCCFR with Pluribus negative-regret pruning, VR-MCCFR control-variate baselines, and OCHS 8-quantile L1/EMD hand clustering. Cuts exact turn/river BR exploitability by 42.5%.
TurnValueNetTorch (2,664 → 256 → 256 → 2,652) exported to binary VN01 + Burch safety-gadget continual resolving bounding mean BR margin to 1.7 chips (<0.1% pot) and cutting max combo exploit by 86.5%.
Quantifies 12 core leaks in bb/100 and ranks corrective drills via hybrid token/3-gram TF-IDF + EV-loss severity (evaluated with NDCG@k, MRR, and Brier score), paired with solver-grounded AI coaching.
| Capability | Static Cloud Viewers | AI Poker Lab |
|---|---|---|
| Solver Architecture | Pre-solved cloud lookup | O(1) spotHash DB + Native Rust 6-Max/HU Linear MCCFR + Python CFR+ |
| Depth-Limited & Safe Resolving | — | ✓ PyTorch ReBeL Value Net (VN01) + Burch Safety-Gadget (-86.5% max exploit) |
| Explains WHY (blockers, MDF, equity) | — | ✓ Streaming AI Coach grounded in solver JSON (temp 0.3) |
| Leak Detection & Drill Ranking | Basic stats | ✓ 12 EV-loss detectors + TF-IDF / severity IR drill recommender (NDCG@k) |
| Benchmark & Statistical Rigor | Closed | ✓ Published BR/LBR, 24k Cross-Play, AIVAT & Welch t-test benchmarks |
6 positions × 4 stack depths (20bb, 40bb, 100bb, 200bb). Inspect raise, call, and fold frequencies plus hand EV in real time.
AI Poker Lab is architected by Daniel Sawa (BSc Computer Science, University of York). Every component—from the multi-threaded Rust 6-Max/HU Linear MCCFR solver (engine/noregrets), PyTorch belief-state value network (value_net_torch.py), and vectorized NumPy CFR+ subgame solver (showdown_solver.py) to the Next.js 15 coaching & IR drill engine—is built for mathematical transparency and ToS-compliant post-hand study.
Solver-backed guides by Daniel Sawa: GTO fundamentals, preflop mastery, MDF & equity realization, and tool comparisons.
Nash equilibrium, why solvers fold 30% on Q72r, and how to study GTO without memorizing 10k charts.
AI PokerWhy post-hand training with a CFR solver beats real-time bots — and keeps your bankroll safe.
Preflop StrategyHow to read range grids, why BTN opens 48% and UTG only 15%, and how to drill ranges until automatic.
24 interactive charts (6 positions × 4 stacks) — hover for frequencies and EV. Validated O(1) pre-computed GTO ranges.
Join 2,000+ players improving with AI coaching. No spam, unsubscribe anytime.
Open the Lab, paste a hand, and inspect your leaks in seconds.
Go to Dashboard