Client project · Industrial
Industrial knowledge base
Half a million pages, answered in seconds.
For an oil and gas company, Ctrl+Space Labs turned half a million pages of an industrial site’s technical and procurement documentation into a knowledge base that engineers query in plain language, with every answer linked to the section it came from.
Oil and gas company · name withheld
- 500,000pages of technical and procurement documents
- ~15 saverage time to the first answer
- 20+employees testing with their own questions
The problem
Six minutes to look one thing up.
Finding a single fact in a document archive takes about six minutes by hand. Done ten times a day, that is an hour of every engineer’s day spent reading instead of working.
By hand~6 minutes, estimated
With the knowledge base~15 seconds, measured
Built with Gendox
What the knowledge base does.
Search the whole archive
One question runs across half a million pages, including single project files of more than 80,000 pages.
Read technical language
Relevant sections are found in seconds, even when the question uses the site’s own terminology.
Show the source
Every answer links to the exact document section it came from, so engineers can verify before they act.
Runs in the client’s environment
Scanned pages were OCR’d on GPU servers and the platform was deployed inside the client’s own environment.
Delivery
From document dump to supervised pilot.
The proof of concept ran in five phases, with 15 weeks from document processing through to the end of the supervised pilot.
- 01
Requirements
Real scenarios, success metrics and a representative document sample.
- 02
Document processing
3 weeks
Around 500,000 pages OCR’d, digitized, classified and quality-checked.
- 03
Customization
4 weeks
Projects per use case, tuned models, extraction rules and reports.
- 04
Testing and rollout
4 weeks
Acceptance testing with employees, training, and deployment.
- 05
Supervised pilot
4 weeks
Real-world use, feedback, monitoring and final adjustments.
Results
Benchmarked on the client’s own scenarios.
The client defined 42 real scenarios and its employees wrote hundreds of questions to test them. The product as configured reached 72% accuracy across those scenarios, and the gaps pointed to four specific areas to improve.
72%accuracy across 42 scenarios
Accuracy = pass ÷ (pass + answers with errors), across all 42 scenarios.
- 42 client-defined scenarios
- Hundreds of employee questions
- 4 improvement areas identified
Contact
Get in touch.
Tell us what you are building. We will tell you whether we can help.