Client work · AI architecture
Library-backed extraction.Every answer is one you know.
Overview
Two retrieval paths run side by side: one reads the document, the other searches a library of risks and controls. Extraction is bound to that library, so every answer is a control the reviewers already recognise, or nothing at all.
- the document and the library
- 2 paths
- every answer maps to the library
- Bound
- standardized, or built with you
- Any domain
The model
It works in standardized domains, where that library already exists, and in unstandardized ones, where we build the library with the client first and the same pipeline runs on it.
- Reads the document
- Searches the library
- Extracts, bound to it
- A reviewer signs
We design AI frameworks and architectures like this for other needs too: agent workflows with a person in the loop, document review, decision support. Each one is shaped around the job, and runs where the data already lives.
Stack
Retrieval
- Dual RAG
- Library-backed extraction
- Reranking
Libraries
- Standard control libraries
- Custom libraries built with the client
Workflow
- Human-in-the-loop review
- Agentic workflow engine
All work
Hello from Malmö
Adamass
Come with the work that feels heavy. We will make it light.
- IKEA
- Nasdaq
- Tele2
- Thomson Reuters
- The Guardian
- Liverpool FC
- Stabenfeldt
- Eyeradar
