SMART Platform Client-Driven Feature Enhancements

Role: API/ML Engineer & Project Coordinator | Client: OU DISC (SMART crisis-intel team) | Tools: Python, FastAPI, GitHub

Real-world scenario illustrating SMART’s analyst workflow during the Presidential Inauguration in 2017.

Real-world scenario illustrating SMART’s analyst workflow during the Presidential Inauguration in 2017.

Misinformation Model System Architecture.

Misinformation Model System Architecture.

Auto-generated endpoints and schemas that sped up client testing and onboarding.

Auto-generated endpoints and schemas that sped up client testing and onboarding.

Quick Links: Poster

Executive Summary:

Partnered with the SMART (Social Media Analytics & Report Tool) client team to make the platform more trustworthy and maintainable for analysts working disasters and large events. We delivered two client-requested upgrades:

  1. Personalized misinformation models (per-user branching from a base model) so analysts’ labels actually stick and predictions reflect their judgment

  2. Django to FastAPI migration of the relevance-classification API for faster response, cleaner docs, and easier upkeep.

Client & Engagement:

  • Discovery & scoping: Ran working sessions with the client lead to clarify pain points (shared model disagreement, heavy API), define acceptance criteria, and prioritize the backlog.

  • Governance: Bi-weekly client checkpoints (demo with Q&A), weekly internal stand-ups, and a living Kanban board (backlog, in progress, review, done).

  • Change management: Captured change requests (Python version constraints, endpoint parity), sized them, and re-ordered sprints with the client’s sign-off.

  • Handover: Delivered API docs, runbook, and quick-start test scripts.

What We Built:

1. Multi User Misinformation Model Support

  • Design: Load a base model at first login; on first relabel, spawn a user-specific model; route all predictions/updates to that user model.

  • Outcome: Increased analyst trust and usability; different users aren’t forced into one “average” model.

2. API Modernization

  • Mapped 4 legacy endpoints to minimal FastAPI routes.

  • Moved metadata to JSON with in-memory retrieval.

  • Resolved Python 3.5 dependency issues.

Project Management Highlights

  • Backlog to user stories with acceptance criteria; risk & decision logs.

  • Ran 2-week sprints with measurable demo increments.

  • Handover package: runbook, endpoint docs, and quick-start test scripts.

Impact

  • Selected approach for SMART 2.0 (client adoption).

  • Higher analyst confidence via personalized models.

  • Faster iteration & easier maintenance from a lean, documented API surface, reducing average start up time by 80%.