About
Built for the people who have to sign it off
Why it exists
The evidence usually exists. Finding it is the expensive part.
Compliance, quality and risk teams answer the same kinds of question every cycle. Does this supplier hold the certificate we require? Was the calibration done? Did the board approve the statement? The answer is usually somewhere in a folder of contracts, certificates, registers and policies, written in words that are not the words of the question.
Keyword search misses it, and general-purpose chat tools answer confidently whether or not the document says so. DataStruct was built for the middle ground: find the passage that actually supports the requirement, show exactly where it came from, keep it apart from material that merely looks relevant, and say so when there is nothing to find.
How we work
Four principles, applied to the product and to this site
Evidence first
Every finding links to the passage it rests on: the file, the page and section, or the sheet and rows. A finding without a source is not a finding.
The reviewer decides
In compliance work a person is accountable for the judgement, so the tool never records one. It finds, cites and sorts; a named reviewer confirms, rejects or asks for more.
Say no plainly
When nothing supports a requirement, the finding says so. The same rule runs through this website: where the answer to a security question is no, we write no.
Measure before claiming
Capabilities are judged against a fixed internal benchmark, not by impression. Where we measured something and it did not work, the roadmap says so.
What it runs on
The stack, for the person doing due diligence
- FastAPI and Python
- Application backend
- Next.js and TypeScript
- Web application
- PostgreSQL with pgvector
- Text, passages and search vectors
- Hybrid search
- Keyword, semantic and section signals merged
- OpenAI
- The only AI provider that receives content
- Celery workers and Redis
- Background processing and caching
- Cloudflare R2
- Original files, scoped per workspace