Documents classified and their data in the system, with no retyping
A company with a lot of technical and product documentation already had a management system, but classifying documents and entering their data was manual work. Now a webapp archives them, classifies them, moves their data into the management system and answers questions citing the document each answer comes from.
The challenge
Hundreds of documents arrive every month, often dozens of pages long. The management system was there, but someone had to open each document, work out what type it was and retype its data by hand. Slow, repetitive work, every day.
Architecture: from upload to the management system
The AI makes the first pass on every document. When a case is doubtful, a person decides, from the same webapp.
What made it work
- One archive with search: documents are uploaded from the webapp and kept in one place, where a search finds them again.
- Automatic classification: each document is recognised by type, without anyone having to open it.
- Data extracted into the management system: the fields that used to be retyped by hand arrive already filled in.
- Answers with a source: people ask questions about the documents and each answer cites the document it comes from, so it can be checked at once.
- A person on doubtful cases: when the AI is not sure, an operator reviews the document and corrects what needs correcting.
- Clear roles: I built the Angular webapp on my own. The Python backend and the AI services on AWS were built together with other developers.
Outcome
Delivered in 2 months instead of the 4 quoted. The company handles hundreds of documents a month, often dozens of pages long, with hours of manual work saved every day. People no longer retype data: they check the doubtful cases.