Future Star AI Agent
A multi-agent assistant for youth baseball
I designed a guarded, multi-agent information assistant that helps families quickly find camp schedules, fees, and answers without searching across a website.
Making the complex understandable.
Parents needed fast, accurate answers, but schedules, fees, and policies lived across different content areas. The system also had to resist prompt injection, avoid exposing private information, and stay inside a narrow youth-sports support scope.
Direction with hands-on depth.
I architected the orchestration model, classifier schema, global state, intent routing, retrieval strategy, guardrails, and end-to-end conversational experience. I also built and tested the live beta as part of a larger site rebrand.
Designed to work beyond one screen.
A strict JSON classifier routes each request to a specialized schedule, fee, or FAQ agent. Each agent retrieves from isolated, curated knowledge, while global safeguards handle out-of-scope requests and PII. The result is a simpler front door to complex information.
Evidence that the work moved things forward.
- Approximately 30% lower latency after workflow optimization
- All out-of-scope prompts in the tested prompt set were safely contained
- Dates and fees remained accurate across the tested knowledge set
- Reusable architecture for local sports and catalog-based assistants
From thinking
to shipped work.
Process, systems, and final executions from the original case study.
