Gressingham Foods Automation Toolkit

January 2026 – Present. Since this is employer work, specifics stay generic — no proprietary data, and no exact volumes or figures unless confirmed safe to share.

Problem

A food production environment with legacy tooling (tkinter-based internal apps) and manual, error-prone compliance processes — including UK Extended Producer Responsibility (EPR) packaging classification and shelf-life data management — running on locked-down hardware with no cloud access.

Approach

Lead with the OCR pipeline — it's the strongest technical story.

Built and deployed a fully local, CPU-only LLM-based OCR pipeline using llama.cpp, running on locked-down hardware with no GPU and no external network access
Automated shelf-life data merging, PDF classification, and Outlook-based reminders for compliance deadlines

TODO: what documents the OCR pipeline processes and why OCR was needed — calibration certificates? Supplier paperwork?

TODO: how accuracy was validated / what the fallback was for low-confidence extractions

TODO: migrated legacy tkinter tooling to Flask — why? Maintainability, multi-user access, web deployment across sites?

Outcome

TODO: time saved, error reduction, or adoption — even an estimate like "cut manual document processing from X hours to Y" is far more compelling than a feature list

Deep involvement in FoodOps/QualityApp administration, including writing conditional expressions and debugging live data sources.

The constraint that makes this notable: getting an LLM pipeline running entirely offline on locked-down hardware is a genuinely uncommon skill to demonstrate.