Lychee: An Enhanced Chess Puzzle Trainer

Lychee is a free and open-source web service that intelligently categorises and recommends from a large selection of puzzles, employing

• known spaced repetition techniques, and
• provision of related puzzles, according to our novel puzzle similarity metric.

Lychee aims to optimise chess tactical training for serious enthusiasts, additionally boasting

• in-depth progress analytics, and
• user-customisable training sessions.

Lychee is heavily inspired by (and largely relies on) the free and open-source Lichess server and its contributions. It was designed and built by myself and five other students at Imperial College London for the Software Engineering Group Project module.

Specifics

This project provides the following.

• A refinement of Lichess’ puzzle tagger that produces more specific, custom, hierarchical tags. These are then used by our puzzle similarity algorithms.
• A Leitner System for fine-grained spaced repetition that overcomes the noted shortcomings of repetitively reviewing the same puzzle, by using highly similar ones instead (according to our metric).
• Coarse-grained spaced repetition, at the level of puzzle themes, balanced with our Leitner system to combat its overfitting.
  • We employ the SM-2 algorithm with collected user data as quality metrics, such as continually maintained per-theme Glicko ratings and time taken to solve puzzles.
• A comprehensive analytics dashboard that displays user progress for each theme (over time), which feeds into a theme trainer mode that allows users to focus on specific groups of themes and switch between them.