Why us
Most AI pilots in compliance fail for the same reason.
They are built by people who have never had to defend a determination. The output reads plausibly, cites nothing traceable, and collapses the moment an assessor or a regulator asks how a conclusion was reached.
We came at it from the other direction. DASATECH is an ISO/IEC 17020 accredited assessment firm that got tired of the reading, matching and drafting hours, so we built AssessIQ to remove them, without ever letting the software conclude on an assessor’s behalf.
That constraint is the hard part, and it is the part we know how to build. Where the determination has to remain with a qualified human, and the reasoning has to be traceable to source evidence, the architecture looks very different from a chatbot with a document uploader.
- A production platform running on live engagements, not a prototype
- Deep domain expertise in the frameworks the tool has to satisfy
- Multi-cloud AI experience across Vertex AI, Bedrock and Azure AI Foundry
- Tenant isolation, audit trails and spend guardrails built for regulated work
- An accredited assessment practice that has to live with the output
Design principles
Five rules we build to, learned the expensive way.
The human concludes
Software drafts, flags and assembles. A qualified person decides, and the record shows who decided and when.
Every claim is traceable
Output cites the source page it came from. If a reviewer cannot follow a statement back to evidence, the statement should not exist.
Human edits are sacred
Refreshed model output lands beside a person’s words, never over them. A tool that silently overwrites expert judgment gets abandoned.
Data stays where you put it
The authorized data path is chosen per tenant and per engagement, and content stays inside it on every call.
Adversarial QC, not spot checks
A second pass argues with the first on every item. Where the two disagree, the question goes to a human rather than being resolved quietly.
Cost is a control
Per-tenant spend guardrails with alert and hard-stop thresholds. A guardrail pause never changes a result.
What we build
Where document-heavy expert work is the bottleneck.
Classification, extraction and synthesis across large mixed corpora (PDFs, spreadsheets, transcripts and images), mapped to a structured requirement set with citations back to source pages.
First-draft generation in a defined house style, with review queues, sampling, comment and resubmission cycles, and locked human edits.
Cross-source conflict detection, dual-pass adversarial checking, and escalation paths that default to human adjudication.
Structured exports to Word, Excel and machine-readable formats such as OSCAL, built from signed work rather than assembled by hand.
Tenant isolation, role and permission models, SSO integration, tamper-evident audit trails and authorized AI data paths across Vertex AI, Bedrock and Azure AI Foundry.
The other direction: independent assessment of AI systems your organization has built or bought, against the control baseline you answer to.
Engagement model
Three ways this usually starts.
- License AssessIQ. If you run assessments, the platform already exists. Fastest path to the hours back.
- Build on the same foundations. We adapt the architecture behind AssessIQ to a different document-heavy expert workflow in your domain.
- Advisory and assessment. We review an AI system you are building or buying, covering architecture, data paths, controls and the human-in-the-loop design, before it becomes a finding.
Tell us what the contract requires. We’ll tell you what it takes.
A 30-minute scoping call is usually enough to size the gap, name the deliverables and give you a realistic date for authorization.