Mistake-based chess puzzles are drills generated from the positions you actually got wrong in your own games — not a random corpus. Chess Masti AI automatically converts your biggest blunders into puzzles, then matches them against a 100,000-puzzle Neo4j graph by 49-dimensional FEN cosine similarity so you drill the geometry of your specific mistake, not a generic theme bucket. Free, engine-validated, no subscription.
Standard puzzle apps give you a randomized sequence: a back-rank mate from someone else's game, a Russian endgame from 1985. The patterns are real but they aren't your patterns. Your blunders cluster — most players lose to the same handful of mistakes on repeat.
Drilling the position your brain just failed to see is the highest-leverage repetition you can do, because the pattern is fresh and the mistake is yours. Drilling random tactics from someone else's corpus is fine; drilling your own is faster.
Stockfish flags the blunder
Every move you played gets a centipawn-loss score. The moves with the biggest evaluation drop are the candidates — that's where the lesson is.
The position one move before becomes the puzzle
The puzzle prompt is the board state right before your mistake. Stockfish's preferred move is the solution. You see the position; you solve it without the answer visible.
Similarity search finds 3 nearby positions
The Neo4j graph of 100,000+ Lichess puzzles gets queried for puzzles in your skill band and the relevant tactical theme, then re-ranked by 49-dimensional FEN cosine similarity. The output: three puzzles whose geometry matches the one you just got wrong.
Drill them inline
Puzzles render directly inside the coaching message — same chat bubble, same chess board. No tab switch, no flow break. Solve, miss, solve again. The SM-2 spaced-repetition scheduler files the position away for later review.
This is not a replacement for tactics training on a curated set. Chess Masti's mistake-puzzle set is built from your real games — if you only play 10 games a month, you only have 10 games' worth of mistakes to drill. For raw volume, pair it with Lichess puzzles, Chessable, or Chess Tempo.
Also worth flagging: not every mistake makes a clean puzzle. Strategic errors and positional drift don't convert to single-solution drills the way tactical blunders do. The puzzle generator skips those (correctly).
See the bigger pipeline
Mistake puzzles are the training-loop half of the free AI chess coach. The other half is the engine-first explanation pipeline. The main page covers how Stockfish, Claude, and the validator fit together.
Open the free AI chess coach pageTurn your mistakes into puzzles
Paste a PGN. The biggest blunders become drills automatically. Free.