P.I.M.P / The AI Division / RAG Systems
The Archive
RAG Systems
An answer without a source is a rumour with better grammar.
Request the Sit-Down01 / The Problem
Four hundred thousand documents across a file server, a wiki nobody updates, and eleven years of mail.
Search returns everything and answers nothing, so people ask the one colleague who still remembers — and wait until she is back from holiday.
The knowledge is not missing. It is unreachable, which costs exactly the same.
02 / What We Do
The Work
- 01
Ingestion that survives your file server
PDFs, scans, spreadsheets, mail, wiki, ticket history. Including the ones that were scanned crooked in 2014.
- 02
Citations, always
Every answer names the document and the passage it came from. If it cannot cite, it says so instead of inventing one.
- 03
Permissions inherited, not reinvented
The system sees what the person sees, enforced against the source system. A retrieval layer that leaks your HR folder is worse than no system at all.
- 04
A test set with a number on it
Real questions from your own people, with correct answers, run on every change. Quality becomes measurable instead of a feeling.
- 05
Freshness
Reindexing on a schedule you choose, so last week’s price list does not keep answering this week’s questions.
03 / How It Runs
The Run
Assessment
1–2 weeksWhat exists, what is readable, what is contradictory. We tell you if the data cannot carry it.
Build
4–8 weeksPipeline, retrieval, permissions, interface. Measured against the test set from day one.
Run
ongoingReindexing, monitoring, and a quarterly review of the questions people ask but the archive cannot answer.
04 / Also On This Block