PawnParse
AI for Chess Sheets
Introduction
PawnParse is a web-based tool that transforms handwritten chess score sheets into fully structured PGN files.
Built with computer vision and machine learning, it allows users to upload photos of game records and instantly get digitized, editable versions of their chess matches.
Target audience: chess players, coaches, tournament organizers, and analysts.
The Challenge
Manually digitizing handwritten chess score sheets is time-consuming and error-prone. The ML model in PawnParse handles two core tasks:
Detecting bounding boxes around move entries
Reading the handwritten text within those boxes
The product needed an interface that:
Allows users to review and correct predictions with minimal friction
Displays confidence scores clearly
Guides the user through the full process from scan to export without overwhelming them
Objectives
Step-by-Step Guidance
Breaks the digitization process into clear, intuitive stages — from upload to PGN export.
Minimal Cognitive Load
Each screen is focused and simple, helping users stay oriented and avoid overwhelm.
Transparent AI Feedback
Confidence levels and corrections are clearly visible, so users always know what to trust or fix.
Smooth Export Flow
Finalized games can be quickly saved or sent to online platforms in PGN format.
Insights
Through close collaboration with the client, we discovered that:
Users prefer step-by-step flowsrather than simultaneous panels
Displaying the original image and digital result side by side increases clarity
The backend operates in asynchronous phases, so users need real-time progress and status updates
Manual intervention is often required, especially when prediction confidence is low
UX Solution
Instead of a static 3-panel layout, we implemented a multi-step linear process, optimized for clarity and control.
Upload & File Dashboard
Users upload one or more score sheets. Each file entry shows live processing status — from uploading, to AI prediction in progress, to ready for review or export.
Bounding Box Correction
Users are shown all predicted bounding boxes on the original sheet. They can manually adjust, add, or delete boxes. This ensures the AI reads the correct regions.
Data Reading
The backend recognizes the handwritten text within each box. A horizontal progress tracker informs the user how far the task has gone.
Label Correction
Predicted moves are editable. Users can see confidence ratings as star-based indicators and correct any mistakes. Metadata (players, date, location) can also be reviewed and edited.
Game Review & Export
A final summary view presents all moves in table format next to the original image. From here, users can export the PGN file or send it to Lichess.
Key Features
PawnParse simplifies a complex technical process into a clear and intuitive user journey. Below are the core features that support users from the moment they upload a photo to the final game export.
Editable Bounding Boxes
Users can add, drag, resize, or delete bounding boxes directly over the scanned sheet to help the AI read the right regions.
AI Confidence Ratings
Each move prediction is paired with a confidence score, shown visually to help users prioritize corrections.
Move-by-Move Editing
Predicted moves are displayed as editable input fields, making it easy to correct any mistakes in context.
Metadata Input
Users can review and update game metadata like player names, event date, and location.
Step-Based Workflow
The app uses a clean, multi-step process to guide users through upload, correction, and export — with progress indicators at every stage.
PGN Export
Corrected games can be downloaded in PGN format or exported directly to chess platforms like Lichess.
Outcome
This UX revision resulted in:
Clearer structure and improved usability across all phases
Drastically reduced onboarding time for new users
Higher completion rates for full game digitization
A polished MVP ready for stakeholder demos and future investment
Alexander Lind
Founder of PawnParse
My Role
UX audit and restructuring of initial concept
Designing the step-based interface architecture
Creating high-fidelity mockups in Figma
UI optimization for asynchronous feedback
Advising on visual design, interaction states, and user education
