Omni - Text to API

Employees spend over 21 hours every year on menial tasks. This replaces some of the data capture tasks with an easy to use API.

Comment

Inspiration

Whenever I am doing research on a specific topic, I open 20-30 pages and go through them looking for specific things that I want to find. I always end up creating a google sheet to keep track of things.

What it does

It takes in raw pages as input and provides you a way to extract entities out of the raw form text and return it in a JSON format specified by you.

How we built it

  • Used Google Cloud Functions for the serverless infra and API endpoint
  • Built a chrome extension to download raw text from a webpage
  • Used claude2 to convert raw text into JSON format

Challenges we ran into

  • I started with using saved html files but that was too much content
  • I looked into embeddings to solve the problem of auto categorization, but we are not there yet
  • Ran into the challenge of getting just a JSON as output

Accomplishments that we're proud of

  • Building all these things within the short time period
  • Keeping the tech stack slim
  • Keeping the user interactions to as minimal as possible

What we learned

  • How to enforce model to return JSON outputs
  • How to use embeddings

What's next for Omni - Text to API

  • Auto capture pages and put them in cloud storage
  • Background tasks to process data async
  • Semantic search using ScaNN over embeddings

Built With

Share this project:

Updates