Inspiration
We wanted to create a new experience for League of Legends, beyond what you'll see in widely used sites such as op.gg. Most stat trackers give numbers, which you have to analyze yourself. Replay.gg was born with the idea of combining riot data with AI coaching so players can get personal feedback.
What it does
Replay.gg analyzes your recent games through the Riot API, calculates detailed statistics (win rate, KDA, most-played champions, etc.), and uses AWS Bedrock (Claude 3 Sonnet) to generate dynamic coaching summaries. Each summary gives tailored insights for your last 5, 10, and 15 matches helping users understand their short term and long-term performance. It also provides AI-generated champion suggestions, recommending three new champions to master based on your current playstyle and recent trends.
How we built it
Frontend: Built in Next.js 16 (React 19) using the new App Router and styled entirely with Tailwind CSS.
Backend: A Node.js + Express API that integrates directly with the Riot Developer API for player and match data.
AI Layer: The backend sends aggregated game statistics to AWS Bedrock, where Claude 3 Sonnet generates structured JSON summaries and champion recommendations.
Challenges we ran into
Some of the challenges we ran into were Riot API limitations. Since we do not have a production key, the amount of data we can pull for each user was limited. On top of that, we had to constantly update the riot key every 1-2 days, and the data was also not the most recent.
Team challenges fell into play here as well, with team members constantly juggling between interviews/assessments/hackathons and other non career related activities, the team size went from 5 to 3, reducing our time frame in the last 1-2 weeks.
Accomplishments that we're proud of
I am still proud of what we were able to accomplish as a team; a fully ufnctional full stack app with AI driven insights with real time Riot data. Also implemented multi segment coaching summaries for 5/10/15 games in a single Bedrock call, reducing latency. Also learned how to merge data analytics and generative AI into a player focused product.
What we learned
A lot of us gained hands on experience with AWS and how powerful AWS Bedrock can be for producing insights from structured data. How to structure AI prompts to return strict JSON that can be parsed safely in production. How to optimize API usage by slicing local data instead of making redundant external calls.
What's next for Replay.gg
Production Riot API key integration for persistent live data. Caching layer (Redis or DynamoDB) to reduce Riot API calls and enable player history tracking. Deeper AI analysis: champion synergy suggestions, suggestions on how to improve cs/min or gold/min. Deployment on Vercel + Render for a permanent public demo. Expanding “Rift Wrapped” into a shareable annual report players can post to social media.
Log in or sign up for Devpost to join the conversation.