Dashcam Library

What needed solving

A day of dual-channel dashcam recording produces roughly 500 clips and 490GB of footage, GPS telemetry in a proprietary format no general tool reads, and a camera clock that disagrees with GPS.

The client needed to find any moment by time and place, with no cloud services involved at any point.

What I owned

• Wrote the scope and acceptance criteria for a three-milestone fixed-price engagement
• Built the Python/FastAPI backend, the GPS telemetry parser and the trip engine
• Built the React interface and the FFmpeg-based player
• Tested every acceptance criterion against the client's own footage

Never guess, never silently correct

Every recording keeps GPS time, camera time, filename time, container time and filesystem time side by side, with a defined fallback order and the source of the resolved time always shown. Duplicate or missing GPS seconds are preserved as-is; invalid fixes are flagged, not dropped.

Files are identified by content fingerprint, not path: a clip renamed or moved to another drive is still the same file, and the same clip filed twice is stored once.

A trip engine that rebuilds instead of reprocessing

Trips are derived from stored metadata and GPS, grouped by a configurable gap threshold, and rebuilt automatically when that threshold changes. Every extraction run is versioned, so a better parser later can re-decode the raw telemetry already stored with each clip without touching the video files again.

What it's built with

Backend

PythonFastAPISQLite

Frontend

React

Media

FFmpeg