Nederlands

AI document and email processing

Your team reads the same kinds of documents every day and types what matters into another system. I build software that does the reading and the extraction, so people only check and decide.

Tell me about your project

The problem

Applications, invoices, reports and customer emails rarely follow one format. Someone opens each one, finds the relevant facts and copies them somewhere else. It's slow, it's error-prone, and it doesn't scale when the volume grows.

Language models are very good at this kind of reading, but a chatbot on its own isn't a process. You need the output in a fixed structure, checks on every field, an audit trail and an easy way for a person to correct the result.

How I approach it

  1. Start with real documents

    We collect a representative set of documents, including the messy ones, and agree on exactly which fields you need and what counts as correct.

  2. Extract into a fixed structure

    A language model reads each document and fills a defined schema: names, amounts, dates, products, each with a reference to where in the document it came from.

  3. Check before anything moves on

    Rules validate every field: formats, totals, required values and cross-checks against your own data. Anything uncertain goes to a person instead of straight through.

  4. Fit into the tools you already use

    Results go to your CRM, ERP, SharePoint or database, or to a simple review screen. Your team doesn't have to learn a new platform.

  5. Measure and improve

    We track accuracy on a fixed test set, so a change to a prompt or a model is measured instead of guessed.

What to watch out for

  • If the documents contain personal data, the model doesn't need to know who the person is. I pseudonymise the text before it goes to a model.
  • Scans and photos need good OCR first. The extraction can only be as good as the text it receives.
  • Don't let a language model make the final decision. Use it to gather the facts, and let fixed rules or a person decide.
  • The unusual documents take most of the effort. The process needs a clean path to a person for anything that doesn't fit.

Projects like this

First assessment of applications

Reading and checking every application by hand takes reviewers a lot of time. The system reads the application file, checks it against external registers and prepares a first assessment for the reviewer. The language models never see data that identifies a person.

  • Personal data is replaced with pseudonyms before any model sees it, and every re-identification is logged.
  • The decision comes from a fixed set of rules, kept separate from the language model, so an auditor can inspect them.
  • Reviewers adjust the rules themselves, with a full history of every change.
  1. Application fileExternal registers
  2. Pseudonymise
  3. Extract facts
  4. Apply the rules
  5. First assessment for the reviewer

Automatic news monitoring for a client list

Searching the news for each client by hand doesn't scale. A scheduled job collects articles every night, matches them to the client list and classifies them. In the morning, the team sees only what matters, such as an insolvency.

  • Every night, the job pulls in articles from many news sources.
  • It flags important events and filters out routine news.
  • The team reads one short list each morning instead of searching by hand.
  1. News sources, each night
  2. Match to the client list
  3. Classify the event
  4. Morning list of important items

Frequently asked questions

Which documents can you process?

Anything with text in it: emails, PDFs, Word files, scans and web pages. Scans and photos go through OCR first. Consistent formats make a project faster, but they aren't required.

How accurate is it?

That depends on the documents and the fields, so we measure it on your own examples before anything goes live. When the system isn't sure, it says so and a person checks the result.

Can it run without sending data to OpenAI or a similar provider?

Yes. Depending on the volume and the privacy requirements, the models can run on your own servers or with an EU provider, and personal data can be pseudonymised first.