SynergyScope

Inspiration

The inspiration came from observing how esports teams struggle to understand performance patterns across game patches and meta shifts. Players often lack data-driven insights into their synergy with teammates and adaptation speed, relying instead on intuition. This platform addresses the gap between raw match data and actionable competitive intelligence.

What it does

SynergyScope analyzes League of Legends match data through multiple AI agents to quantify player synergy, measure adaptation speed to meta changes, and predict optimal team compositions. The system processes match history via Riot API, builds relationship graphs in Neptune, trains GNN models for synergy analysis and generates natural language insights through Bedrock.

SynergyScope employs a multi-agent system architecture with seven specialized agents: Social Graph Agent: Builds player relationship graphs using AWS Neptune Chemistry Analyst Agent: Quantifies synergy using SageMaker GNN models Meta Analyst Agent: Tracks patch changes using Glue and Athena Adaptation Agent: Models learning curves using SageMaker Forecast Compatibility Agent: Recommends compositions using SageMaker and Bedrock Storyteller Agent: Generates insights using Bedrock Visualizer Agent: Creates dashboards using QuickSight and Amplify

How we built it

The platform uses AWS Lambda for API ingestion, Glue for ETL processing, Neptune for graph storage and SageMaker for training GNN and time-series models. Bedrock generates narrative insights from structured data, while QuickSight and a React frontend visualize synergy networks, adaptation heatmaps and performance predictions.

Challenges we ran into

Correlating performance changes with specific patch notes required complex temporal analysis across multiple data sources. Training GNN models to detect subtle synergy patterns between player pairs demanded extensive feature engineering. Translating dense statistical outputs into actionable recommendations while maintaining technical accuracy presented significant design challenges.

Accomplishments that we're proud of

Successfully implemented a multi-agent architecture where each AI component specializes in different analytical tasks and shares insights. Built a working system that quantifies intangible concepts like player chemistry and meta adaptation speed. Created a pipeline that processes raw match data into predictive recommendations for future team compositions.

What we learned

Graph neural networks excel at modeling relationship-based patterns that traditional ML approaches miss. Time-series forecasting combined with graph analysis provides deeper insights than either method alone. AWS serverless architecture enables cost-effective processing of sporadic analytical workloads.

What's next for SynergyScope

Expand support to additional games like Valorant and Dota 2 by generalizing the agent architecture. Implement real-time analysis during live matches to provide in-game strategic recommendations. Add psychometric profiling to link playstyle traits with synergy patterns and develop a coaching feedback loop that tracks recommendation adoption rates.

Built With

• amazon-web-services
• amplify
• api
• api-gateway
• athena
• bedrock
• bootstrap
• chart.js
• cloudwatch
• css3
• d3.js
• deepar
• docker
• dynamodb
• error
• eventbridge
• fastapi
• forecasting
• functions
• gateway
• geometric
• glue
• gnn-models
• graph-neural-networks
• handling
• hosting
• html5
• javascript
• lambda
• neptune
• pydantic
• python
• pytorch
• quicksight
• react
• s3
• sagemaker
• sagemaker-recommender
• scripts
• series
• step
• terraform
• three.js
• time