Synapse
Human-Centered AI Assurance Framework — YorkU × TD Bank capstone scoring Jira user stories against versioned AI/privacy policies in real time.
Stack: Python · FastAPI · TypeScript · NestJS · Next.js · MongoDB · Auth0 · Atlassian Forge
Timeline: Sep 2025 – April 2026 · Team of 7 · Client: TD Bank
Overview
Synapse is a Human-Centered AI Assurance Framework built for TD Bank as my York University capstone project. It scores Jira user stories in real time against a versioned AI / privacy policy framework, surfacing violations and remediation hints inline via an Atlassian Forge plugin so policy review happens where work is already being done — not after the fact.
What I built
- Risk-scoring microservice — Python / FastAPI service powered by OpenAI Structured Outputs and a hierarchical policy schema. Returns deterministic, auditable risk assessments per user story.
- Policy management API — NestJS + MongoDB backend with full CRUD, activation workflows, revision history, and Auth0-based RBAC.
- Admin console — Next.js front-end for policy authors to draft, version, and activate policies; deployed to Render.
- Forge plugin — Embeds the assessment results directly in Jira issue views with severity-coded badges and remediation hints.
Collaboration
Worked in a multidisciplinary team of 7 (engineering + UX + product) under real client constraints from TD Bank, balancing security, auditability, and developer ergonomics.