System diagram
Co-founder project · Document AI
Project Parse
Clean JSON from invoices, with a Gemini fine-tuning loop.
- AI
Proof: Demo
Co-founder project · Construction
Estimate AI
Construction estimates turned from messy PDFs into structured, ERP-ready data.
System diagram
- 01
Vendor PDFs
Quotes, scopes, estimates
- 02
Examples
Input and output pairs
- 03
Fine-tuning
One model per vendor
- 04
Output
Structured, ERP-ready data
- Input or existing system
- We built
- Built by
- Muhammad Zubair, Sortup co-founder
- Sector
- Construction
- Stack
- Django
- React
- OpenAI fine-tuning
- PDF parsing
- Celery
- Redis
- Client
- A construction estimating team handling vendor quotes in many formats.
- Engagement
- Zubair built the Django and React application and the fine-tuning workflow behind it.
The challenge
Quotes, scopes and estimates arrive as PDFs that differ by vendor. Manual extraction is slow, error-prone and hard to connect to ERP or estimating tools.
The solution
Customers create a model per vendor, upload input and output examples, prepare datasets and run fine-tuning jobs, then use the trained model to extract structured output from new documents. PDF parsing and job scheduling run on Celery and Redis.
The impact
- A separate model for each vendor format.
- Training examples come from the team’s own documents.
- Structured output ready for downstream systems.
Building something like Estimate AI?
No pitch deck. An engineer, and a straight answer.