Trinity: Agentic AI-powered transition planning for students with disabilities
Amazon Web Services Sadia Ahmed
Trinity helps disabled students build their own transition plans instead of filling out static forms. It’s now scaling across U.S. districts and expanding to Saudi Arabia and Kuwait.
Based on reporting by Amazon Web Services, Sadia Ahmed — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
University Startups has pushed its Trinity tool from a promising idea into a production system for students with disabilities, and the hard part was never the chat interface. The challenge was taking a process that has to be legally aligned, accessible, and specific to each student, then making it work as more districts signed on and the original setup stopped being enough.
The answer was to split the work. Trinity moved from one model trying to do everything at once to a six-agent system on AWS. An orchestrator handles the flow, while separate agents focus on colleges, jobs, training programs, community options, and independent living goals. That matters because each part of the process pulls from different kinds of records, and a mistake in one area can poison the whole session.
The recommendation side is grounded in three Amazon Bedrock Knowledge Bases. College matches draw from more than 30,000 records, employment from more than 1,000 occupations, and training from more than 5,000 programs. Trinity also mixes semantic similarity, keyword matching, location preference, and program attributes like disability support, cost, and education level, then drops low-relevance results entirely. The system is built to keep answers tied to actual records rather than whatever the model feels like saying that day.
Accessibility and compliance were treated as core infrastructure, not cleanup work. Student data is protected with field-level encryption using AWS Key Management Service, access is limited by role, and login runs through Canvas LTI 1.3 instead of a separate credential system. Audio readouts via Amazon Polly and speech-to-text through Amazon Transcribe mean the backend doesn’t have to change just because the student needs a different way to interact.
Trinity is already reaching educators and students in more than a dozen U.S. states, and international rollout into Saudi Arabia and Kuwait is underway. The pitch is simple: give students ownership, give educators speed, and do it without turning special education into a compliance circus. That’s the right instinct. Too much school software still acts like the student is an afterthought and the staff are the customers; Trinity at least tries to reverse that.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI education product that sounds like it was built by people who actually read the rules before shipping. The multi-agent setup is not glamorous, but in special education, boring and correct beats clever every time. More vendors should try that instead of stapling a chatbot onto a compliance problem and calling it innovation.
Read more about this at: Amazon Web Services
Related stories
Trinity: An Evolved LLM Coordinator
Sakana AI ·
30
How AI tools can redefine universal design to increase accessibility
Google Research · 6 months ago ·
29
How Together AI Uses AI Agents to Automate Complex Engineering Tasks: Lessons from Developing Efficient LLM Inference Systems
Together AI · 1 year ago ·
6