Back to Projects
ai/ml
2026-Present

DigiPop

Receipt-intelligence platform that intercepts USB print traffic and runs multilingual OCR and LLM extraction.

Key Metrics

7

Repositories

7

Languages Extracted

MVP in progress

Status

About This Project

DigiPop is a 7-repository receipt-intelligence platform I lead as architect with a team of 3 engineers. It intercepts USB print traffic between POS terminals and thermal printers, parses ESC/POS into structured receipts, and runs cloud OCR and LLM extraction for merchant analytics. The investor demo was delivered, and the production MVP is now being built toward a 30-merchant Philippines pilot.

I reverse-engineered a black-box vendor USB man-in-the-middle device (Allwinner H3) end to end with no documentation, and discovered and wrote the mitigation runbook for a four-channel supply-chain backdoor. I built a streaming ESC/POS parser (a byte-by-byte FSM over raw USB capture) and a two-stage multilingual extraction pipeline (RapidOCR plus Azure OpenAI, 7 languages) with no-hallucinate prompting.

Technologies Used

Python FastAPI RapidOCR Azure OpenAI ESC/POS Embedded Linux PostGIS Docker

Interested in Similar Work?

I'm available for freelance projects and full-time opportunities. Let's discuss how I can help bring your ideas to life.

Get in Touch