Pixie

Inspiration

As college students, we're constantly taking pictures or screenshotting campus events, internship opportunities, club meetings, and important deadlines, only for them to get buried in our camera rolls and never to be seen again.

We wanted to bridge the gap between saving information and actually using it. That's how we came up with Pixie — a little magic for your camera roll!

What it does

Pixie is an AI-powered photo organizer that transforms forgotten pics into organized, actionable information.

Simply upload a picture, and Pixie uses AI to identify important details like event titles, dates, locations, deadlines, and tasks. Instead of manually organizing everything, users can:

Automatically categorize screenshots into Campus Events, Career Opportunities, Extracurriculars, Personal, and more.
Review and edit AI-extracted information before saving.
Track events and deadlines through a personalized dashboard and interactive calendar.
Manage actionable tasks associated with saved screenshots.
Search, filter, and browse an organized camera roll.

Our goal is simple: turn passive screenshots into opportunities you can actually act on.

How we built it

We built Pixie as a full-stack application, combining React Native, Expo, and TypeScript for the frontend with Python and FastAPI for the backend.

Our screenshot-processing pipeline uses Tesseract OCR to extract text from images, followed by Qwen 2.5 running locally through Ollama to interpret that text, categorize screenshots, and identify relevant information. We also use Pydantic for data validation and SQLite for storage.

We designed five interconnected screens: Dashboard, Camera Roll, Confirmation, Item Details, and Calendar, creating a seamless experience from uploading a screenshot to organizing its information. Throughout development, we used GitHub for collaboration amongst teammates.

One of our main priorities was exploring local-first AI, allowing us to process information without relying on cloud-hosted language models.

Challenges we ran into

Building Pixie came with several challenges:

Unstructured information: Pictures vary significantly in layout, formatting, and content, making accurate extraction difficult.
AI reliability: We needed to handle missing or inaccurate information without blindly trusting model predictions.
Frontend/backend integration: Developing components in parallel required careful coordination of API endpoints and data structures, ensuring that the various components would merge smoothly.

Accomplishments that we're proud of

We're especially proud of transforming a problem we personally experience into a product we'd genuinely want to use.

We created an application that brings together picture organization, event tracking, task management, and an interactive calendar. We also designed a human-centered AI workflow that allows users to review and correct extracted information rather than simply accepting it.

Beyond the application itself, we're proud of exploring locally running AI and collaborating as a two-person team to bring our idea to life.

What we learned

Building Pixie taught us that creating an AI-powered product involves much more than integrating a language model. We learned how to connect OCR, local LLMs, backend APIs, and frontend interfaces into a cohesive application.

We also gained hands-on experience with React Native, Expo, FastAPI, and Ollama, while strengthening our skills in GitHub collaboration, API design, and debugging.

Most importantly, we learned the value of human-centered AI: making AI useful means designing systems that are not only intelligent, but also transparent, reliable, and easy to use.

What's next for Pixie

We see Pixie evolving into an even more intelligent personal organizer. Some features we'd love to introduce include:

Automatic camera roll scanning to organize screenshots without manual uploads.
Google Calendar and Apple Calendar integration for seamless event scheduling.
Personalized recommendations based on users' interests and priorities.
On-device mobile inference for greater privacy and convenience.

Ultimately, we want Pixie to make sure that important information doesn't get lost just because it's saved in a screenshot.