Landing Page
Behavioral Leaks cards
Analysing Page
Personal Dashboard
Roadmap and Performance Chart
Statistics

Inspiration

As a state-level chess player, I have spent years playing competitive chess and practicing regularly through online platforms. While tools like Stockfish and platform analysis are excellent at identifying mistakes in individual games, I often found myself repeating the same types of errors despite reviewing hundreds of games.

The problem was that traditional analysis focuses on positions, not players.

It can tell you what move was wrong, but it cannot explain why the same mistake keeps appearing across different games. After studying my own games, I noticed recurring behavioral patterns such as rushing under time pressure, becoming complacent after winning material, or relaxing immediately after castling.

That realization inspired ChessLeak — a tool that analyzes the player rather than the position and uncovers the hidden habits that repeatedly cost games.

What it does

ChessLeak is an AI-powered chess coaching platform that analyzes a player's game history and identifies recurring behavioral weaknesses, or "leaks."

Instead of reviewing a single game, ChessLeak studies hundreds of games to discover patterns such as:

Time Collapse: Performance dropping dramatically under time pressure.
Post-Castling Relaxation: Increased mistake rates immediately after castling.
Complacency: Failing to convert winning positions after gaining material.
Opening Dependence: Over-relying on openings that consistently underperform.

The platform ranks these leaks by severity, generates a personalized chess persona, and provides actionable coaching advice along with a structured improvement plan.

How we built it

We built ChessLeak using:

Python and Flask for the backend
Pandas for large-scale game analysis
python-chess for PGN parsing and board-state evaluation
Lichess Public API for retrieving player game histories
OpenAI GPT-4o-mini for generating coaching insights and behavioral diagnoses
HTML, CSS, JavaScript, and Chart.js for the frontend experience

The workflow is:

User enters a Lichess username.
The system fetches up to 200 recent games.
Chess metrics and behavioral indicators are extracted.
Statistical analysis identifies recurring patterns.
An impact score ranks each leak by severity.
AI generates a human-like coaching report.
The user receives a complete diagnosis and improvement plan.

Challenges we ran into

One of the biggest challenges was moving beyond traditional chess analysis. Finding behavioral patterns across hundreds of games required designing metrics that were statistically meaningful rather than anecdotal. We had to ensure that every insight was backed by sufficient sample size and measurable evidence. Another challenge was translating raw chess data into understandable coaching advice. Engine evaluations and numerical statistics are difficult for many players to interpret, so we used AI to transform complex data into clear and actionable recommendations.

We also had to balance technical depth with usability, ensuring the report felt like a coaching session rather than a complicated analytics dashboard.

Accomplishments that we're proud of

Successfully shifted the focus from game analysis to player analysis.
Built a system that identifies behavioral patterns rather than isolated mistakes.
Created an AI-generated coaching report that feels personalized and actionable.
Designed a ranking system that prioritizes the most impactful weaknesses.
Combined chess analytics, behavioral psychology, and AI into a single user experience.
Built a complete end-to-end MVP capable of analyzing hundreds of games in minutes. Most importantly, we created a tool that solves a problem we personally experienced as competitive chess players. --- ## What we learned Building ChessLeak taught us that improving at chess is not only about finding stronger moves but also about understanding recurring habits and decision-making patterns. We learned that behavioral insights become much more valuable when supported by data, and that AI can play a powerful role in converting technical analytics into coaching advice that users can actually act on. We also discovered how difficult it is to balance statistical rigor with simplicity, especially when presenting insights to players with different skill levels. --- ## What's next for ChessLeak Our next goal is to evolve ChessLeak from a diagnostic tool into a complete improvement platform. future features include: l Integration with Chess.com game histories, m More advanced behavioral metrics and psychological profiling, m Personalized training plans based on detected weaknesses, m Weekly progress tracking and improvement reports, m Opening-specific coaching recommendations, m AI-generated training exercises tailored to each player's leaks, m Comparative analysis against players of similar rating levels, t Ultimately our vision is to create the first AI chess coach that doesn't just analyze games but truly understands the player behind them.