LEO AI / Document AI
Handwritten and scanned academic documents, turned into clean text.
An AI transcription platform for two US-based co-founders that reads handwritten and scanned documents and returns formatted, summarised text.
Large documents processed without blocking the appAsync
Real-time job status for every uploadLive
Built as a senior engineer at Algotix AI · Visit tryleo.ai ↗
The challenge
OCR on handwriting is messy, and processing large documents is slow. Users needed accurate results without the app freezing while jobs ran.
What was built
- OCR followed by GPT clean-up to reconstruct context, fix formatting and summarise.
- Heavy jobs run in a Bull queue off the main thread, with live job status in the UI.
- Job creation and status logic in Postgres functions for consistency.
- Drag-and-drop uploads, Stripe payments and Sentry monitoring for launch.
Stack
Next.jsClerkNestJSSupabase / PostgresBull queuesStripeVercel + Railway
Get a build like this
The matching packages.
More work
Creator tools↗
< 15 minTime to publish, down from hours (reported)Owli.ai
From hours to under 15 minutes to publish a video everywhere.
Education / Voice AI↗
< 500msVoice latency target for live conversation (reported)Teacher's Pet
A real-time AI voice tutor that adapts to how each child learns.
Automotive↗
+220%User engagementDealer-360i / BuyBidCars
A multi-tenant dealership ERP that lifted user engagement by 220%.
Have a problem like this?
Tell me what is slowing you down. I will reply with the clearest next step.