Inspiration
We wanted to help midwives digitize paper registries through a familiar conversation, while keeping them in control.
What it does
Oneday turns registry photos into structured records. Midwives can confirm, correct, or retake photos, manage multipage documents, search patient files, and enter data manually when AI is unavailable.
How we built it
We used Python, FastAPI, OpenCV, RapidOCR, and vision-language models, with a JavaScript interface. Records use encrypted local storage and a queue for simulated offline processing and synchronization.
Challenges we ran into
Reading handwriting, distinguishing blank from illegible fields, preserving unanswered questions, and grouping pages correctly were our biggest challenges.
Accomplishments that we're proud of
We built a complete capture-to-verification workflow with patient matching and recovery from failures. On 80 clean synthetic pages, our local pipeline achieved 99.93% correctness among automatically accepted fields.
What we learned
Useful AI must explain uncertainty and offer clear next steps. Testing the actual interface was essential to finding conversation and upload bugs.
What's next for Oneday
Test with midwives, evaluate real and Arabic handwriting, support more registries, and develop production offline synchronization and WhatsApp integration.
Log in or sign up for Devpost to join the conversation.