The gap is not “more generative AI”
Raiqon’s founders previously built Codebeamer, a requirements platform that is now common in automotive ALM. That history shows up in the problem they open with: regulated product development already has tools (DOORS Next, Codebeamer, Polarion and peers). What teams lack is assistance that is enterprise-ready - security and certifications, behind-the-firewall deploy, workflow integration, cost control at scale, and determinism (same input, same output on Monday and on Friday).
Szabi framed the contrast with API wrappers clearly. Individual copilots can make one engineer faster. Raiqon aims at whole workflows that stay audit-ready and repeatable - funded across many OEMs and tier-1s so the industry invests once in a shared product rather than twelve fragile in-house RAG stacks.
What “holistic” means on the V-model
They position the platform across the engineering V: requirement improvement, project-level consistency (duplicates, similarity, contradictions), impact and traceability, decomposition from customer requirements toward system requirements with reuse of existing items, and test-case generation - including homogeneous test sets rather than one-off cases.
Scale is not theoretical. Their largest cited customer sits above 30 million requirements (tests on top). Cost was part of the same story: roughly on the order of 1/150 of always calling a frontier model such as GPT-5, via task-appropriate models among roughly two dozen domain-specific ones, on-prem or private cloud.
Demo: from one requirement to project health
Marc’s live demo stayed inside the host ALM (Codebeamer in the session): single-requirement improvement against standards and INCOSE-style formality; project tracker health (issues, duplicates); semantic traceability checks with graph suggestions; decomposition with reuse; then agent and workflow builders that write back into the host system so teams do not export-edit-reimport.
If your bottleneck is regulated requirements and tests, the useful question is not “which chatbot?” It is whether assistance is inside the system of record, deterministic enough for audit, and cheap enough at tens of millions of objects.