Trinity: how a disability student-planning AI scaled from one prompt to a six-agent Bedrock architecture
AWS ML Blog · rss · 2026-09-03
AWS ML Blog profiles Trinity, a conversational AI by University Startups that helps students with disabilities build IDEA-aligned IEP transition plans. Within a year it reached educators and students across a dozen+ US states, with Saudi Arabia and Kuwait expansion underway.
From one model to six agents: Early prototypes crammed intake, career exploration, compliance checking, and plan writing into a single prompt, causing hallucinations. The rebuilt architecture uses an orchestrator plus five specialist agents (College: 30k+ schools; Employment: 1k+ occupations; Training: 5k+ programs; Community; Independent Living).
Key designs:
- Two-phase pipeline: 5-10 ranked recommendations first, then a consolidated IEP plan exportable to PDF in 5-10 seconds
- Hybrid retrieval scoring: semantic 30%, keyword 25%, location 20%, program attributes 25%, with a 25% relevance cutoff
- HIPAA/FERPA compliance built in: field-level encryption, role-based access, retention controls
- Accessibility via Polly text-to-speech, Transcribe speech input, and Bedrock Guardrails
More from coding & agent
- One-person company in 2026 = you plus four Grok bots running marketing — PrajwalTomar_ · 2026-09-03
- Rival AI agents: cross-vendor model review catches what self-review misses — rseroter · 2026-09-03
- Web Draw: MCP server reads pages as text instead of screenshots, Amazon page ~750 tokens — ahstanin · 2026-09-03
- Let Claude write its own /compact prompt and follow-up message — zsakib_ · 2026-09-03
- VibeCAD + McMaster parts search? Early impressions say Fable 5.1 is quite good — burhop · 2026-09-03
- Claude Code Opus 5 Auto Mode hijacked via prompt injection with up to 80% success rate — bibryam · 2026-09-03