Your renewal payment of
£240.00 is due
by
30 November 2026.
Extract → Review → Confirm
September 2026 to Present
LifeAdmin AI
Evidence-grounded document intelligence
Finding the actions, amounts and deadlines buried in everyday documents to with evidence people can check.
Evidence before confidence.
Engineering notes / Problem, decisions & progress
Individual engineering & research project
The problem
Bills, invoices, tenancy documents and renewal notices can bury an important action inside lengthy text or a complex layout. A plausible extraction is not enough when a wrong date could matter.
My contribution
I’m designing and evaluating an evidence-grounded extraction system as my individual final-year Engineering Project.
Technical approach
The planned comparison uses OCR-and-rules, a multimodal-model baseline and a hybrid pipeline. All three will return the same structured fields, linked to source evidence, so results can be compared fairly.
Engineering challenges
Distinguishing a date mentioned in a document from a real deadline, handling OCR errors, and rejecting unsupported model output. The review workflow is designed to let users correct or reject fields before confirming a reminder.
Result & current status
Currently in development. The evaluation plan uses versioned synthetic documents and held-out layouts, measuring field accuracy, precision/recall/F1, evidence support, false deadlines, processing time and cost. No accuracy results or live demo are claimed.

