Release 0.1 — archived. These are the notes for the previous release. The current release is Release 0.2.
Version 0.1 is the first co-build release of TrueLumen — an AI review assistant that detects verifiable mismatches across engineering project documents, with citations to the exact pages, without replacing engineering judgment. It is early software and is actively improving. Your feedback during the pilot will directly shape its checks, document handling, and review experience.
Each organization has its own isolated workspace. Your team's projects, documents, and results are visible only within your organization. Administrators can approve users, select AI models, oversee verification review, and monitor check health.
Upload engineering PDFs and let TrueLumen infer which files are versions of the same document, which document governs another, and how the set fits together. You confirm or reject every inferred relationship before relying on it — you decide; the tool never decides for you.
Check drawings and other project documents against governing specifications. TrueLumen produces four types of findings:
Every finding includes verbatim source material and citations to the exact PDF pages and sentences.
Compare two design documents as equals — for example, civil and electrical drawings — to surface disagreements between disciplines, without assuming either document governs the other.
Check one document against several specifications in the same run, ranked by authority so TrueLumen knows which governs on conflict. Disagreements without a clear ranking are flagged for an engineer to resolve.
Findings keep stable identities across repeated checks. A judgment recorded once — real issue, immaterial, or error — carries forward automatically, and previously judged findings are collapsed rather than raised again. For every finding you can record a verdict, its usefulness (actionable / good to know / noise), its project impact (critical / major / minor / none), and the reviewer's name — including typed names for reviewers without accounts.
Each project keeps a history of its checks so the team can revisit previous results.
Ask questions about your uploaded documents; answers are grounded in those documents. Conversations are saved to your account and remain private to you — project administrators cannot read them in the app.
Send comments or feature requests from inside TrueLumen; we are notified by email.
TrueLumen learns each project’s requirements from its own documents and checks. Early checks deliberately favor precision, and coverage then grows with every check as the project’s requirement knowledge builds.
A single run currently assesses at most 120 requirements; requirements beyond that limit are not assessed in that run.
Plan: a configurable limit.
The same underlying issue can appear as more than one finding in cross-document checks.
Plan: de-duplication.
Plan: add redundant AI models so a temporary outage no longer reduces confidence labels or coverage.
TrueLumen may have smaller review coverage on new disciplines, such as civil drawings, producing fewer cross-discipline comparisons than expected.
Plan: TrueLumen already learns each project's drawing conventions from its checks, and coverage builds across runs. The planned improvement moves this learning earlier: learned extraction rules will be activated at file-upload time rather than at check time, so full coverage is ready before the first check. Engineers can review and revoke any learned rule at any time.
A check interrupted during a service restart can incorrectly appear as "canceled by user".
Plan: fix this issue.
A first deep check reads every page and builds the project’s requirement checklist. Recorded runs have taken from a few minutes to about 90 minutes, depending on the document set. Later checks on the same set reuse finished work; in one September 2026 test, a repeat check ran 56% faster.
Plan: optimization work to significantly reduce check times, for both first and repeat checks.
Geometric and spatial clash detection, and support for more document types, are planned for a later release.
Please report confusing results, missing comparisons, document-handling problems, and feature requests through the in-app feedback widget. Co-build feedback is most valuable when it includes the project context, the result you expected, and what TrueLumen showed instead.