—RAG
RAG systems and AI agents
A RAG agent is not a generic chatbot: it ingests your documents, retrieves the right context and runs business logic (calculations, catalogue selection, PDF reports). Everything inside your infrastructure. No sensitive data in third-party clouds if that is what you decide.
Real example
Tank manufacturer: an agent that calculates, selects equipment and generates a PDF report. €39,000 a year saved.
€39,000/year saved
—Architecture
Everything inside your server
Documents → ingestion → vector database → agent → your team. Data sovereignty is not a slogan: it is the perimeter.
- 01Your documents→
- 02Ingestion pipeline→
- 03Vector database→
- 04Agent→
- 05Your team
- Everything inside your server
—Deliverables
What you get
- Ingestion pipeline (PDFs, catalogues, internal databases)
- Vector database on your server (pgvector or equivalent)
- Agent with citations, business logic and report generation
- Interface for your team (web app or integration into the tools you already use)
- Access controls and query traceability
—Fit
Who it makes sense for
Knowledge trapped in two or three people's heads or in unstructured folders
Quotes or technical reports that currently take hours
Sectors with sensitive data that cannot go to a generic LLM SaaS
—Stack
- pgvector
- FastAPI
- Docker
- VPS
—Questions
Frequently asked questions
Does our data go to OpenAI or Anthropic?
It depends on the design. We can use external APIs only with non-sensitive data, or run local or private models. Data sovereignty is defined during the diagnosis.
Does it work with any PDF?
It works with reasonable, well-scoped document sets. Garbage in, garbage out: week one defines which documents go in.
How do I know an answer is reliable?
Citations to the sources, human validation in critical steps, and error metrics during the test phase.
Does it fit your case?
30 minutes. If a SaaS tool is enough, I will tell you. If custom software makes sense, you leave with a fixed scope.
Discuss your case — 30 min