Blackline Forensics

Blackline Forensics

Built with a small team to give investigators a place to upload suspect media and get authenticity scores. I owned the backend scaffolding, wired the model runner, and paired with the front-end to surface inference results cleanly.

Prototyped an ingestion API in FastAPI, packaged the model runtime with Docker, and stitched in a signed S3 flow plus PostgreSQL audit trail so we could safely persist artifacts. The web front-end was kept intentionally simple so the team could focus on reliability and handoff docs.

Key Features

• Hooked a FastAPI ingestion endpoint to a PyTorch model runner with pre/post-processing to keep detection scores stable.:
• Added an audit trail with Postgres plus signed S3 asset storage so client uploads stayed isolated and traceable.:
• Packaged the dev stack with docker-compose and a short handoff guide so the rest of the team could iterate faster.:

Learnings

• Balancing GPU-heavy inference with predictable API latency by shaping payload sizes and timeouts.
• Keeping ML-heavy repos maintainable with typed Pydantic schemas and background task runners.
• Small UX touches (job states, retry affordances) make even technical dashboards feel trustworthy to non-engineers.