Marta Almeida
View profile
Construction-site progress photo tracker
Weekly drone-footage frames → vision model identifies milestones (foundation, framing, roof, drywall) → updates project dashboard.
Project managers caught delays ~3 weeks earlier on average.
0
Problem
Construction firm got weekly drone footage of 14 active sites. Project managers eyeballed footage for progress — late catches meant late penalties.
Flow
Weekly drone footage upload → frame extract at cardinal points → vision model → milestone tagging → dashboard update + drift alerts.
Cadence
Weekly per active site, 14 sites tracked
What I built
- Frame extraction at fixed cardinal points per site
- Vision model identifies construction milestones (foundation → roof)
- Per-site progress timeline plotted against original schedule
- Alert when actual vs scheduled drift exceeds 14 days
Limits
- Weather affects detection — heavy snow/rain weeks are tagged uncertain
- Civil/structural focus — does not assess interior or MEP work
AI tools used
- Replicate
- Claude
- Supabase
@marta.almeida