Founder & Lead Engineer
DotTrim
A production-grade desktop application that helps video editors work faster through AI-powered video indexing and pre-montage assistance. Designed around real-world constraints such as large local media libraries, intermittent connectivity, and editor-specific workflows.
Problem
Video editors spend a significant amount of time manually scrubbing through footage, identifying usable segments, and organizing clips. Pure cloud solutions are often impractical due to file sizes, privacy concerns, and unreliable connectivity.
Solution
Built a hybrid local-first system that performs video processing and indexing on the user’s machine, with optional cloud-based AI processing for heavier analysis. The system generates structured metadata that editors can use to quickly assemble rough cuts and explore their footage.
Architecture & Approach
Desktop application backed by a local FastAPI service written in Python, using FFmpeg and OpenCV for video processing. AI analysis integrates Gemini, WhisperX, and CLIP. Local indexing is stored in SQLite, with PostgreSQL used for cloud-side services. Additional microservices are implemented in Node.js.
Key Tradeoffs
- Local-first processing increases client complexity but avoids large uploads and latency
- Hybrid cloud support adds architectural overhead but enables more advanced AI analysis
- Balancing performance with cross-platform desktop support required careful resource management
What I Learned
- Desktop applications introduce challenges not present in web apps, including packaging, updates, and OS-specific behavior
- Editor workflows vary significantly; early user feedback influenced both UX and system design
- AI pipelines need robust fallbacks when models fail or produce low-confidence results