Product Intern, Jun to Aug 2025
← V Naman Mehta
An AI pipeline for international trade paperwork.
During my internship I worked on a system that reads shipment documents and pulls out the fields a trade team needs. OCR, Azure OpenAI and rule-based checks do the extraction.
- Ingestion
- AI extraction
- Validation
- Report generation
How it runs
Four containerised services
The system runs as four containerised services, each with one job.
Python, Flask, FastAPI, Docker, Azure Functions, Container Apps, Blob Storage
- Ingestion
- Takes the shipment documents into the pipeline.
- AI extraction
- Pulls the fields out of the documents, using OCR and Azure OpenAI.
- Validation
- Checks the extracted fields against rules.
- Report generation
- Turns the results into a report.
Results
Accuracy, volume and turnaround
The numbers from the internship. They are the ones on my resume.
- Extraction accuracy
- 95 to 98%.
- Volume
- 500+ shipment document packages a day.
- Turnaround
- Down from two to three days to under five minutes.
More
Other work, or get in touch
Next: the Secure Browser at Talview.