Your knowledge is in PDF documents. Can you use it?
Getting text out of a document has been solved for years. Understanding what it says has not. We build systems that turn documents into structured data, answers and working products.
Where knowledge gets stuck
In companies the most valuable knowledge usually sits in documents, only findable by whoever already knows where to look.
Getting text out of a document was solved years ago. Understanding what it says wasn’t, and that’s the work: structure, page types, values that have to be right.
Images are the expensive part
Hardly any PDF comes without images: diagrams, scans, embedded tables, photographs. For a model that’s the most expensive part of the job.
Providers now accept PDFs and images directly, which tempts everyone into sending everything. Send only what needs a model, and solve deterministically whatever can be.
Our clients hire us to build what gives them an edge. Confidentiality is part of the job. Publishing the details could give that edge away. We’ve delivered in regulated environments and on projects worth hundreds of millions.
What we've built
An end-to-end pipeline for a client, starting from extensive specialist reports as PDFs: excellent content, digitally almost unusable.
What comes out isn’t a PDF viewer but an app-like web experience, generated in one pass. Where a value has to be exact, a deterministic check against the original overrules the model.
An internal tool inside a client's operation. Handwritten forms are read, validated and turned into clean records.
Where the system can’t read something with confidence, it says so instead of guessing, and only those fields reach a person. A correction made by a person is never overwritten.
Also built by 2x10
Content Agent
specialist literature evaluated automatically, every finding traceable to its source.
Lab parser
lab values read from medical reports into structured data, whatever the layout.
Compliance Checker
content checked against internal guidelines and brand rules, not by sampling.
What makes them hold up
Document work fails on the exceptions, not the average case. Three questions find them early.
- Which mistake is actually costly?Not the empty field but the one read wrongly that nobody noticed, so every extraction is validated against a fixed schema.
- Are two documents ever the same?The pipeline decides how to approach each one instead of forcing a single route onto all of them.
- Where can the LLM model be relied on?In domains with no tolerance for error that needs to be measured and monitored.
What we could build for you
- We turn documents into structured records in your databases, so paper and PDFs become something you can query.
- We find the connections across everything extracted, the ones that were never visible while it sat in separate files.
- We make handwritten notes usable, so what people wrote down actually gets used.
- We surface the obligations buried in contracts nobody re-reads, every finding with the passage it came from.
- And whatever you have in mind that AI and solid software engineering could make possible.
Under the hood
Most engagements start small: talking it through, a few weeks of engineering, something real you can judge. If it works, we take it further.If it doesn’t, you find out early
Data Pipelines
Built to scale
View themeAI Content
Quality at volume
View themePrivate AI
Your data stays
View themeN° 06 Contact
Tell me what you’re planning
No form, no ticket system. Email me directly and you get me, not an inbox someone else manages.
Moritz Miedler · Founder, replies personally