Platform
Grounded Explainability
Every extraction shows its work. Every field traces back to the exact page, paragraph, and coordinates on the source document — with a confidence score, a rationale, and an audit entry. No black boxes, no guesswork.
Source-grounded extraction
Every value tied to where it came from.
Generic OCR returns text. GutenOCR returns text plus its exact origin — the page, the paragraph, the line, the X-Y coordinates. That grounding is captured at extraction time and travels with the data through every downstream step. Reviewers don't search for context. They open the document and the AI shows them.
- GutenOCR coordinate-level grounding. Bevaya's proprietary OCR captures page, paragraph, line, and X-Y position for every extracted token.
- Works on messy real-world documents. Scanned PDFs, handwritten notes, multi-column ACORD forms, faxes, 40-page submission packages — grounded all the same.
- Multi-document indexing. One 40-page packet becomes ACORDs, loss runs, SOVs, and schedules — each component grounded to its own source pages.
- Citations exported with the data. Send extracted fields to Guidewire or your policy admin with the source coordinates attached.
Confidence + rationale on every field
Three numbers, three answers, one glance.
Every extracted field carries three things a reviewer needs: a confidence score (0–1.00), a rationale for how the AI got there, and a citation back to the source. Color-coded so a reviewer can scan a 60-field claim and know in seconds where to focus — and where the AI got it right.
- Field-level confidence scoring. 0.99 green, 0.93 amber, 0.87 routed for review. Thresholds are configurable per field, per flow.
- Plain-language rationale. Not just what was extracted — why. "Date of loss inferred from claim header, cross-referenced with policy effective date."
- Click any field to verify. Opens the source document and highlights the exact location the AI cited. Verification takes seconds, not minutes.
- Verification-layer double-check. A second AI model validates the first. This is how Bevaya gets to 98%+ — and why every score is meaningful.
Governance is the architecture
Audit-ready, by design.
Regulators don't ask whether the AI was accurate — they ask how you can prove it. Every extraction, decision, score, citation, and reviewer action is captured to an immutable log with reviewer identity, timestamp, and original AI output. The same explanation a reviewer sees on day one is the same one your auditor sees three years later.
- Immutable per-action audit log. AI output, reviewer ID, correction value, timestamp, confidence at decision time — logged forever.
- Built for the most regulated insurance markets. Audit-defensible architecture, independently certified, deployed in production at top-tier carriers.
- Exportable for compliance reporting. Pull the full decision chain for any claim, submission, or extraction on demand.
- RBAC + flow versioning. Every change to a flow is versioned. Every user action is attributed. Compliance isn't bolted on — it's the architecture.
Feature index
Inside Grounded Explainability
Every feature that makes a Bevaya AI decision auditable, verifiable, and defensible. Choose a layer to explore.
Source grounding | GutenOCR
The extraction layer. GutenOCR captures source coordinates at the moment of OCR. Every downstream step inherits that grounding — automatically, with no extra configuration.
- Page, paragraph, line, and X-Y coordinate capture
- Scanned PDFs, faxes, and handwritten notes supported
- Multi-column and complex-table layouts
- Multi-document packet splitting, per-component grounding
- Hundreds of carrier-specific ACORD and loss-run formats
- Coordinate metadata exported with the extracted data
- Re-OCR on demand for low-confidence regions
- Real-time document testing during configuration
- Selective recompute — fix one field without rerunning the flow
Verification & review | X-Ray Mode
The reviewer surface. Every field carries a confidence score, a plain-language rationale, and a one-click citation back to the source. Built for adjusters and underwriters, not data scientists.
- X-Ray Mode click-to-cite highlighting
- Visual line connecting field to source location
- Field-level confidence scores (0–1.00)
- Color-coded confidence (green / amber / red)
- Plain-language rationale per field
- Verification-layer double-check on every output
- AI Assistant document Q&A ("where is the date of loss?")
- Doc-ref badges showing which source informed each field
Audit & governance | Compliance
The defensibility layer. Compliance is architecture — not a feature bolted on. Every decision, every change, every reviewer action is captured, attributed, and exportable.
- Immutable per-action audit log
- Capture: AI output, reviewer ID, correction, timestamp, confidence
- Flow versioning — every change traceable
- RBAC partitioning by project and line of business
- Decision-chain reconstruction for any past extraction
- Override-rate metrics per field surface model drift
- Per-field threshold configuration on the Workflow Canvas
- Compliance export ready for regulator and audit reporting
Resources & insights
From the field and from the lab
Research
Page stream segmentation with LLMs
How Bevaya Labs approaches a foundational problem in insurance document AI.
Case Study
Workers' comp carrier processes claims 100x faster
How indexing automation delivered 432% ROI in 12 months.
Architecture
Inside the Bevaya platform architecture
How specialized models, HITL controls, and integrations come together in production.
More Capabilities
Explore the rest of the platform.
Designed, deployed, and governed together. Powered by InsurGPT™ and accessed through the AI Assistant.
Workflow Canvas
Visual builder and production runtime for every automation.
Current page ReviewHuman-in-the-Loop
Configurable review queues with X-Ray verification and a patented feedback loop.
Current page DocumentsDocument Intelligence
Read any insurance document — hundreds of carrier formats, scanned or digital.
Current pageGrounded Explainability
Every value traceable to its source. X-Ray Highlight Mode brings citations to reviewers.
Current pageAnalytics Dashboard
Live accuracy, STP rates, reviewer SLA, and agent performance across every workflow.
Current page GovernanceGoverned Automation
Immutable audit trails, role-based access, flow versioning. Compliance is the architecture.
Current pageTrust & Security
Trust by design
Built for an industry where data security isn't optional.
Your data stays yours.
Never shared with other customers or vendors. Bevaya doesn't train shared models on your data.
Visit the Trust CenterYour data · only your team sees it
Encrypted end-to-end.
256-bit AES encryption, in transit and at rest. Independent third-party audits conducted annually.
Visit the Trust CenterAudited annually · independent third party
Runs in Azure.
Enterprise-grade infrastructure, hosted where insurance organizations already trust their data.
Visit the Trust CenterDeploy where your stack already lives
Every decision audited.
Immutable audit logs. Confidence scoring. Human-in-the-Loop review on low-confidence items.
Visit the Trust CenterImmutable trail · every decision, every reviewer
FAQ
Common questions
Every extracted value is tied — at the moment of extraction — to its exact origin on the source document: the page, the paragraph, the line, and the X-Y coordinate range. That grounding metadata travels with the data through every downstream step, so any reviewer, auditor, or system can ask "where did that come from?" and get an answer in one click. This is built on Bevaya's proprietary GutenOCR technology and is what makes the AI's work verifiable instead of opaque.
Generic OCR returns text strings. Generic IDP returns structured fields. Neither tells you where on the page the value came from, or how confident the model is, or whether a second model agreed. Bevaya's Document Intelligence Layer pairs GutenOCR-level grounding with field-level confidence, plain-language rationale, and a verification-layer double-check — and it persists all of that to an immutable audit log. The result is an AI decision an auditor can defend three years later.
The platform is independently audited and operates in production at three of the top five P&C carriers and across heavily regulated workers' comp, commercial, and specialty lines. Every reviewer action — the original AI output, the reviewer identity, the correction, the timestamp, the confidence at decision time — is logged immutably and exportable for compliance reporting. Your security and compliance teams can review the full certifications list during evaluation.
They're routed to the Human-in-the-Loop review queue along with any business-rule exceptions (missing documents, risk appetite, duplicate flags). Downstream systems — your policy admin, claims system, Guidewire, anything connected to the flow — never see them until a reviewer has resolved them. The reviewer sees the same confidence score, rationale, and citation a downstream consumer would see, plus X-Ray Mode to verify against the source in seconds.
Yes. When Bevaya delivers extracted fields to Guidewire, your policy admin, or any downstream system via API or SFTP, the source coordinates, confidence score, and decision metadata travel with the payload. Your team and your auditors retain full traceability inside their own systems, not just inside Bevaya.
Grounded Explainability is the evidence layer — confidence scores, source citations, rationale, and the audit trail behind every AI decision. Human-in-the-Loop is the review experience that uses that evidence to route low-confidence work to your team. Explainability is what makes a decision auditable whether or not a human reviews it. HITL is what happens when a human needs to step in. Both are built on the same underlying grounding.
GET STARTED
Ready to design, deploy, and govern
your AI workforce?
Bevaya AI Agents can help you triage, analyze, and recommend across underwriting, claims, and policy servicing.
Let's connect and show you how it works.