Inspiration
Our inspiration came from two recent breakthroughs.
The science: fascinating research from institutions like Stanford University has shown that when given accurate personality profiles, LLMs can accurately mimic a human being. Studies have also specifically linked gaming patterns in titles like League of Legends to real-world psychological traits.
The technology: the explosive arrival of powerful Large Language Models (LLMs) like Claude 3.5 Sonnet has given us the ability to not just analyze data, but to reason with it and create natural, human-like conversations. We saw a unique opportunity at the intersection of this science and technology. We asked: What if we could use a state-of-the-art LLM to interpret the proven science behind gameplay data? From this question, Shape Split was born, an effort to create the first true "digital twin" for gamers.
What it does
Shape Split is a web application that creates interactive, AI-powered digital twins of League of Legends summoners. It transforms a player's raw match history into deep, actionable insights about their personality, playstyle, and decision-making.
Here's what you can do: Generate Your Digital Twin: Enter any Riot ID to create a comprehensive profile based on validated psychometric models. See your scores on the Big Five (OCEAN) personality traits and discover which of the 12 Jungian archetypes your playstyle embodies.
Chat With Any Player's Twin: Our core feature. You can have a conversation with an AI that thinks and talks based on a player's specific data.
Personal Coaching: Ask your own twin why you tilt or how to better close out games.
Learn from the Best: Create a digital twin of Faker and ask for advice on mid-lane macro.
Scout Your Opponents: Generate a twin of your ranked opponent and ask them about their greatest weaknesses before the match even starts.
AI-Powered Duo Compatibility: Shape Split analyzes the psychological and strategic synergy between you and a potential duo partner. It goes beyond simple role compatibility to predict how your personalities will mesh under pressure, providing a compatibility score and actionable advice.
Shareable Insights: Every profile and analysis is generated as a clean, shareable card, perfect for posting on social media to compare with friends.
How we built it
Shape Split is a full-stack application built with a modern, serverless architecture optimized for performance and scalability.
Frontend: The client is a responsive web app built with React and TypeScript, ensuring a type-safe and component-driven structure. We used React Router for navigation and the Context API for global state management.
Backend: We opted for a serverless architecture using Node.js functions hosted on Vercel. This allows the application to scale automatically with demand without managing server infrastructure.
Data Pipeline & Analysis: The journey begins when a user enters a Riot ID. Our Vercel backend hits the Riot Games API to fetch up to 100 recent matches. This raw data is processed by our custom personality engine, which calculates dozens of statistical features (like KDA variance, objective focus, and champion diversity). These features are then mapped to the Big Five personality traits and used to determine the closest Jungian Archetype.
The AI Core: The magic happens with AWS Bedrock. The generated personality profile and key statistics are formatted into a detailed context-building prompt. This context is fed to Claude 3.5 Sonnet, which powers the interactive chat, duo analysis, and all other natural language insights.
Challenges we ran into
Effective Prompt Engineering: Getting the AI to be a "digital twin" is harder than just asking it to "act like a gamer." We spent significant time engineering the system prompt, experimenting with how to best inject the player's personality data, stats, and conversation history to get responses that felt authentic, insightful, and consistent with their profile.
Making Data Meaningful: We had a mountain of data, but the real challenge was presenting it in a way that was digestible, visually appealing, and genuinely useful. We focused on creating a clean UI that guides the user from high-level insights (like their archetype) down to the specific numbers that prove it.
Accomplishments that we're proud of
Moving Beyond Stats: We successfully created a tool that provides qualitative, psychological insights rather than just another quantitative stats dashboard. The ability to "talk" to a player's data is, we believe, a true innovation.
The Interactive Twin Feature: We are incredibly proud of the chat functionality. The AI's ability to provide personalized, data-driven coaching by referencing a user's specific traits ("Your high Extraversion makes you a great initiator...") feels like a glimpse into the future of player improvement.
Full-Stack Serverless Implementation: We successfully designed and deployed a complex application on a modern serverless stack. The entire data pipeline, from Riot API to our personality engine to the AWS Bedrock integration, works seamlessly.
What we learned
The LLM is only as good as its context: We learned that the "magic" of an LLM is unlocked by the quality of the data you provide it. Our biggest breakthroughs came not from changing the model, but from refining the personality and statistical context we fed into the prompt for each interaction.
The Importance of API-First Design: Building around the constraints and capabilities of the Riot and AWS APIs forced us to be disciplined and efficient in our architecture.
User Experience is King: Complex data science and AI are useless if the user can't understand them. This project reinforced the lesson that design and user experience are just as important as the underlying technology.
What's next for Shape Split
The digital twin concept has massive potential, and we're just scratching the surface.
Full Team Synergy Analysis: Expand the duo compatibility feature to analyze a full 5-player roster, identifying potential personality clashes and strategic strengths for competitive teams.
Long-Term Personality Tracking: Allow users to track how their gameplay personality evolves over a season. Are they becoming more disciplined? More aggressive? Provide a timeline of their psychological growth.
Pro-Scene and Coaching Tools: Develop a dedicated dashboard for esports coaches to create and manage digital twins of their entire roster, enabling deeper analysis and personalized training regimes.
Log in or sign up for Devpost to join the conversation.