Knowledge assistant
One chat assistant turns scattered documentation into a sourced answer
Technical education client · Machine maintenance
Before a technician could act on a fault, they first had to guess which manual, electrical schematic or system held the answer. Now they ask one plain-language question and get an answer with the exact source attached.
Our approach
The choice was to train nothing on the client's own documentation and instead retrieve the right passage per question, so every answer keeps its file and page and a new manual can simply be added. A question is first classified by type, then the approved documents and the exact figures from the database are collected separately, and only then does the language model write within that bounded material. An unknown fault is stopped rather than expanded into a plausible diagnosis.
The technician gets a sourced answer with the file and page it came from, never a guess.
Before Cognima
- Ask around or search separate manuals and systems one at a time
- Try to match the right document to the fault by trial and error
With Cognima
- Ask one plain-language question in the chat
- The assistant selects the right source itself, every time
- Delivery model
- Chat application over existing manuals and data
- Human oversight
- Technician stays in control; no bypass of safety steps
- Data & IP
- Client-owned manuals, schematics and event data