180 developers. 14,000+ wiki pages. 23 hardware component variants. A search experience so broken that finding answers took longer than writing new code. This is how Wissenswerk turned documentation chaos into a competitive edge.
MedVision Systems is a Berlin-based medical device manufacturer specializing in diagnostic imaging equipment — MRI machines, CT scanners, and ultrasound systems. Their software teams develop the firmware and control systems that run on 23 different hardware component variants, each with version-specific compatibility requirements.
With 180 developers across firmware, application software, and quality assurance, MedVision operates under strict regulatory standards (IEC 62304, FDA 21 CFR Part 11). Every decision, every change, every compatibility matrix must be documented — and findable.
MedVision had a documentation problem that was hiding in plain sight:
Internal surveys revealed that developers spent an average of 18 minutes per search query in the wiki — a task that should have taken 4 minutes. With 34 searches per developer per week, this added up to 6,200 person-hours per year spent just looking for information that already existed somewhere in the wiki.
The company wiki had grown organically over 9 years. Pages were duplicated, outdated, or contradictory. 3,200 pages hadn't been updated in over 2 years. Different teams used different naming conventions — the firmware team called it "HW v3.2 compat," while QA referenced "Hardware Revision 3.2 compatibility matrix." Same document, impossible to find both.
During an internal audit, the QA team discovered that 4 firmware compatibility matrices in the wiki conflicted with each other. Two referenced an outdated hardware revision that had been recalled. In a regulated medical device environment, shipping software based on wrong documentation could result in patient safety issues — and regulatory consequences worth millions.
MedVision's Head of Engineering, Thomas Weber, started with a focused pilot before scaling across the organization:
The 28-person firmware team was chosen as the pilot group. Wissenswerk ingested their 3,400 wiki pages plus 1,200 PDFs from the compliance archive. Within days, the team was using Astra to query firmware compatibility information — and getting accurate answers in seconds instead of 18 minutes.
All 14,000+ wiki pages were indexed, along with 4,800 PDFs, 2,300 architecture diagrams, and 890 meeting transcripts from the shared drive. Wissenswerk's structuring engine unified naming conventions — "HW v3.2 compat" and "Hardware Revision 3.2 compatibility matrix" were automatically linked as the same knowledge entity.
Wissenswerk's conflict detection immediately flagged 247 contradictions in the knowledge base. The most critical: 18 pages referencing recalled hardware revisions and 4 competing firmware compatibility matrices. A dedicated 3-person team resolved the conflicts using Wissenswerk's side-by-side comparison, cleaning up documentation that had been wrong for years.
Senior engineers participated in AI debrief sessions to capture knowledge that had never been written down — why certain PCB layouts were chosen, which vendor chips had undocumented quirks, and what firmware workarounds existed for specific hardware batches. 310 debrief sessions added 1,800 new structured knowledge entries from tacit expertise.
Two incidents demonstrated why Wissenswerk became indispensable:
Wissenswerk's conflict detection flagged that a firmware update scheduled for release referenced a hardware revision that had been internally recalled 8 months earlier. The recall notice existed as a PDF in the compliance folder — but the firmware team had never seen it. Without Wissenswerk linking these documents, the update would have shipped targeting hardware that shouldn't exist in the field.
A new hire on the QA team — who would normally take 3 months to become productive — used Astra to answer 47 questions in her first week. Instead of interrupting senior engineers or reading through hundreds of wiki pages, she got precise, sourced answers about testing protocols, hardware specifications, and regulatory requirements. Her manager reported she was contributing to code reviews by day 12.
After 3 months of full deployment across all teams:
"We had 14,000 wiki pages and nobody could find anything. Developers were re-documenting information that already existed because searching was so painful. Three months after Wissenswerk, the wiki is the same size — but now every piece of knowledge is actually findable. The conflict detection alone probably saved us from a regulatory incident."
Wissenswerk