-
-
user register and login UI
-
Main Page (Recent Tasks and Projects)
-
Project Task Page
-
Upload new task using AI
-
New Task after analysis it
-
Settings page
Inspiration
We often brainstorm on whiteboards but hate manually typing tasks into project boards. We wanted a swift way to snap a photo and instantly get structured tasks including owners, due dates, and professional notes.
What it does
TaskUp takes a photo of a whiteboard, uses Azure Computer Vision for OCR, then Azure OpenAI to parse the text into actionable tasks with owners, due dates, and notes. Optionally, it inserts those tasks into a project database or Trello board for quick collaboration.
How we built it
FastAPI + SQLAlchemy for backend endpoints and data models. Azure Computer Vision for OCR, transforming board images into text. Azure OpenAI to analyze the raw text and produce structured tasks. GitHub Copilot to accelerate coding routines and scaffold endpoints.
Challenges we ran into
Ensuring OCR output is clear enough for AI to parse. Dealing with varied handwriting and partial text. Balancing speed vs. accuracy for rewriting task details. Structuring complex data (teams, projects, tasks) in a single pipeline.
Accomplishments that we're proud of
A fully functioning pipeline: from a messy whiteboard photo to a neat list of tasks with due dates and owners. Integrating multiple Azure services in a short time. Simplifying the user experience—snap a photo, get tasks, done.
What we learned
Multi-modal AI synergy (images + text) is surprisingly effective with Azure Vision + OpenAI. GitHub Copilot let us focus on logic rather than boilerplate, speeding up development. Good data validation (with Pydantic) matters when dealing with incomplete OCR text.
What's next for TaskUp
We plan to:
Refine the AI to handle more complex board layouts and partial handwriting. Add multi-task support so one board image can yield multiple tasks. Extend to voice or video inputs like scanning meeting transcripts to auto-create tasks.