FieldProof
From field photos
to evidence that holds up.
A verification and reporting workflow for NGOs, built around a deceptively simple question: what actually changed?
- Top 25 of ~400 teams
- Code Cubicle 6.0
- Team Compilers
The challenge
Months of field photos end up scattered across phone galleries and chat groups. Dates become uncertain, duplicates pile up, and a donor’s question about progress becomes a manual investigation. FieldProof turns that collection into organised evidence.
A deliberate pipeline
GPS distance identifies the site, EXIF dates help establish the phase, and perceptual hashes flag near-duplicates. Date and location sanity checks keep doubtful photos out of reports. Semantic and keyword search work together, with keyword fallback when the AI provider is unavailable.
Teaching the system to refuse
A convincing description is not proof. The same-location verdict runs before a before-and-after assessment, and a report never cites a pair that failed it. Reports use recorded facts, and each image remains traceable to its original in Cloudinary.
The outcome
Built for Code Cubicle 6.0 and selected in the top 25 from roughly 400 teams. The most important constraint wasn’t generating more narrative—it was making sure the system knew when not to make a claim.
Care about what people see.
Care just as much about what they don’t.
