HexCoreAI

Inspiration

HexCoreAI was sparked by the gap between raw stat trackers and the personal stories players actually want—how their builds impacted fights, how vision shaped objectives, and how their style evolved across a full season. The aim was to create a recap that feels like a coach and a storyteller, turning match logs into a narrative players can reflect on and proudly share.

What it does

HexCoreAI ingests a full year of League MATCH‑V5 data, runs six specialized AI agents (Build, Combat, Vision, Economy, Champion, Competitive) in parallel, and synthesizes personalized insights, highlights, and progress visualizations. It delivers real‑time, trace‑driven updates over WebSocket to a Next.js client and produces a share‑ready year‑end recap with strengths, growth areas, and memorable moments.

How we built it

The system is a serverless, event‑driven pipeline on AWS: API Gateway (WebSocket) connects to TypeScript Lambdas, with SQS and EventBridge decoupling ingestion and processing, DynamoDB storing filtered match shards, and S3 hosting final recap payloads. An Express Step Functions state machine fans out to six Bedrock Agents with Lambda action groups and streaming orchestrators, then aggregates and synthesizes the final narrative for the client.

Challenges we ran into

Balancing Bedrock trace richness with cost required environment‑based switches, progress budgeting, and careful token management for streaming transparency. Handling Riot API rate limits and timeline fidelity meant building robust retry/backoff, selective filtering, and idempotent ingestion without losing key features needed by downstream agents.

Accomplishments that we're proud of

Delivered a real‑time “agent thinking” experience using trace events (rationale, tool invocation, observation) surfaced over WebSocket, making AI analysis tangible for users. Achieved resilient multi‑agent orchestration with partial synthesis, DLQs, and per‑agent error isolation, ensuring engaging recaps even when individual tools or data sources hiccup.

What we learned

How to design Bedrock Agents with clear tool contracts and orchestrators that stream trace events into meaningful UI progress signals users understand. How to map Riot’s timelines into agent‑ready features and correlate cross‑domain insights—build efficiency, combat positioning, vision coverage, economy pacing, champion mastery, and ranked trajectory—into a coherent story.

What's next for HexCoreAI

Add friend leaderboards and duo synergy insights, enabling social comparisons and complementary playstyle recommendations. Expand creative recap formats with generative visuals and templated story “cards,” and introduce longitudinal coaching that tracks targeted goals across split‑to‑split seasons.

Built With

• amazon-dynamodb
• amazon-web-services
• nextjs
• sam
• tailwindcss
• typescript