Built for evidence discovery with human review

Not a chatbot. Not a search engine. A platform for locating evidence across unstructured documents, with citations bound to the claims that use them and human review built in.

Document Intelligence

Hybrid Search

Lexical + semantic + metadata search with reciprocal rank fusion, so evidence is found when the document does not use your vocabulary.

Evidence-Backed Q&A

Ask questions and get answers grounded in your documents. Claims link back to their cited source document, with page and exact supporting span where DataStruct can locate one. Sources it cannot confirm are labelled unverified.

Evidence States

In Founding Customer workspaces each finding separates passages DataStruct could support from sources cited but not confirmed, and from material that is merely related. The reviewer decides; the system records no verdict.

Programme Reruns and History

Re-run an Evidence Programme next quarter without disturbing requirement identity. Earlier findings are retained rather than overwritten, so the evidence position over time is a record, not a snapshot.

Extraction & Analysis

Template Builder

Define extraction schemas with field types, validation rules, and confidence thresholds. Run across any number of documents.

Rules Engine

Deterministic condition types: keyword presence and absence, regex match, numeric thresholds, and missing-disclosure checks. Configurable severity per rule.

Comparison & Benchmarking

Side-by-side document comparison with AI-powered analysis. Build peer groups and benchmark across entities.

Report Generation

Generate HTML, DOCX, or JSON reports with evidence tables, rule results, extraction summaries, and executive overviews.

Enterprise Security

RBAC Permissions

Granular permissions across 4 roles. Control who can upload, search, extract, review, and administer.

Audit Trail

Sensitive actions logged with actor and before/after values: uploads, AI calls, review decisions, rule overrides, configuration changes.

Human-in-the-Loop

Low-confidence outputs auto-route to review queues. Approve, reject, or edit with full decision history.

Evidence Validation

Citation validation, fabricated-reference detection, verification pass, and confidence scoring. Designed to surface unsupported claims so reviewers can act on them.

Platform Architecture

Domain Packs

Pre-built industry configurations covering legal, insurance, healthcare, ESG, and more. Each pack ships its own terminology, document types, extraction templates and rule packs.

API-First

100+ REST API endpoints. Authenticate with JWT or API keys. Build custom integrations and workflows.

SSO — on the roadmap

Google OAuth and generic OIDC sign-on are on the roadmap and are not generally available today. Password and API-key authentication are what ship now.

Webhooks — on the roadmap

Signed webhooks on document events are planned and not available today. The REST API is the integration surface that ships now.