Everything You Need. Nothing You Don't.
Enterprise-grade photo verification with the simplicity of a consumer product. Mobile SDKs, hosted verification, offline retry workflows, and privacy-first design.
Get StartedFast Hosted Verification
Server-managed analysis with mobile-friendly UX.
VerifyAI pairs mobile capture SDKs with a hosted verification API so policies, prompts, and model behavior can evolve server-side without forcing app releases. The result is a fast user flow with centralized control over verification logic.
- No client-side model downloads required
- Policy and prompt updates roll out server-side
- Consistent behavior across iOS, Android, React Native, and Flutter
- Built for in-flow operational verification
Fast Hosted Verification
Semantic Scene Understanding
Understands context, not just pixels.
Unlike simple image classification, VerifyAI understands the full scene context. It identifies sidewalks, bike racks, doorways, parking zones, vehicles, packages, and obstacles. This semantic understanding enables accurate verification even in complex real-world environments.
- Object detection + scene segmentation combined
- Distinguishes between compliant and non-compliant positions
- Works across diverse environments and lighting conditions
- Trained on millions of real-world verification scenarios
Semantic Scene Understanding
Offline Retry Workflows
Keep moving through poor connectivity.
Supported SDK flows can queue transient failures locally and replay them when connectivity returns. That keeps field operations moving in poor-connectivity environments without pretending every capture path is fully offline.
- Queue transient failures locally
- Replay queued work on app resume or manual trigger
- Best fit for base64 verification flows today
- Custom capture pipelines can add deeper offline storage if needed
Offline Retry Workflows
Privacy-First Architecture
Control what you send and how long it lives.
Images are transmitted only when needed for verification, over encrypted channels, with configurable retention and regional storage options. Customers stay in control of submitted metadata, retention policy, and downstream access.
- TLS encryption in transit
- Configurable retention policies
- Regional storage options
- Hosted on SOC 2 / ISO 27001 certified infrastructure
Privacy-First Architecture
Policy-as-Code
Turn any rule into an AI model.
Translate complex verification requirements into executable policies. Define what constitutes compliant parking, acceptable delivery placement, or passing damage inspection—then deploy instantly. Update policies server-side without app updates.
- JSON-based policy definitions
- Per-city, per-zone, per-time-of-day rules
- Update policies without app store releases
- Custom violation categories and severity levels
Policy-as-Code
Real-Time Analytics
Insights, not just data.
Track compliance rates, violation hotspots, fleet health, and operational trends in real time. The VerifyAI dashboard provides actionable insights that help you optimize operations, reduce costs, and improve compliance.
- Live compliance rate tracking
- Geographic violation heatmaps
- Trend analysis and anomaly detection
- Exportable reports for stakeholders
Real-Time Analytics
Continuous Learning
Models improve automatically.
VerifyAI models are continuously retrained on new data to improve accuracy over time. Monthly model updates are pushed automatically—no manual intervention or additional cost. Your verification accuracy improves every month.
- Monthly model retraining included
- Accuracy improvements pushed over-the-air
- No additional cost for model updates
- Custom model training available for Enterprise
Continuous Learning
Enterprise Security
Built for regulated industries.
SOC 2 Type II audit in progress with built-in fraud prevention that detects duplicate photos, AI-generated images, and metadata tampering. Role-based access control, full audit logs, and SSO integration for enterprise compliance requirements.
- SOC 2 Type II audit in progress
- Fraud detection (duplicates, AI-generated, tampering)
- Role-based access control with SSO
- Full audit logs and compliance reporting
Enterprise Security
Damage Intelligence
Per-panel damage with severity, industry-standard codes, repair-cost dollars, and pickup-vs-return delta. Built for insurance, leasing, rental, and finished-vehicle logistics.
Damage detection with severity
Per-panel damage findings with type (dent, scratch, crack, paint chip, missing, misalignment, rust) and severity (none, light, medium, severe). Opt-in per policy via damageMode — existing parking and POD policies are unaffected.
- 11 damage types, 4 severity grades
- Bounding boxes for every finding
- Confidence and panel coverage area per finding
- Cloud VLM today, on-device on the roadmap
AIAG + K1–K5 grading
Damage findings transform automatically into industry-standard grading frameworks. AIAG codes for North American finished-vehicle logistics, K1–K5 for European leasing. Pure server-side transform, no extra API calls.
- AIAG body code coverage (LFD-DT-2, etc.)
- K1 (no damage) through K5 (structural)
- Auto-computed from damage_findings
- Configurable thresholds per customer
Repair-cost estimation
PreviewTurn damage findings into dollar repair estimates via Mitchell Cloud Estimating API. Aggressive server-side caching, fail-soft if Mitchell times out. GT Motive (EU) on roadmap.
- Mitchell parts + labor database
- Per-customer + per-region pricing
- Cached `verify_ai_repair_catalog_cache`
- Verification still returns if vendor is down
Before/after delta detection
Pair a check-out verification with a check-in verification and the engine surfaces only new damage attributable to the rental. IoU>0.3 matching with panel-not-visible exclusion to suppress false positives from differing angles.
- Diff buckets: new, worsened, improved, unchanged
- Vehicle pairing by inspection_session_id or VIN
- Repair-cost estimate on the delta
- Designed for rental + leasing dispute defense
Customer-facing workflows
Capture inspections from end users without forcing an app install. Branded artifacts your customers actually sign and store.
Web-link self-inspection
Send a renter, claimant, or driver a single signed URL. They open it in mobile Safari or Chrome, are walked through each required shot, and sign on completion — no app install. Token-authenticated, HMAC-signed, DB-hashed for revocation.
- Zero-install end-user capture
- Per-customer branding (logo, colors, footer)
- EN / ES / FR / DE locales out of the box
- Resend, expire, revoke from API or dashboard
PDF condition reports
Branded PDFs with annotated damage overlays, vehicle block (VIN / plate / timestamp), and signature panel. Cached to private Supabase storage; signed URL on demand. Renders the damage section automatically when damageMode is on.
- Server-side @react-pdf/renderer
- sharp-composited damage bbox overlays
- Severity-colored annotations
- Inline preview + download link
VIN + license plate OCR
Read VIN and plate from a single photo. ISO-3779 checksum validation, then NHTSA vPIC decode to make / model / year / trim — cached indefinitely. Privacy guardrail: we never look up plates against owner databases.
- Gemini structured-output extraction
- ISO-3779 VIN checksum
- Free NHTSA vPIC decode + cache
- No plate-owner lookups, ever
Enterprise foundation
What enterprise procurement asks for: site-scoped access, exception triage, append-only audit log, claims-platform integrations, and BI export.
Sites + role-based access
Scope verifications to specific sites (yards, dealer lots, depots). Org owners and admins see everything; site managers and viewers only see their assigned sites. Site resolution from metadata.site_external_id or X-Verify-Site-Id.
- Org owner / admin / site manager / site viewer roles
- Nullable site_id — flat orgs keep working
- Per-site KPI dashboards
- Audit-logged role changes
Exception triage queue
Low-confidence and non-compliant verifications open an exception automatically. Assign to a teammate, add notes, mark resolved or dismissed. Every state transition is audit-logged.
- DB trigger auto-opens on confidence<0.85
- Idempotent: UNIQUE on verification_id
- Status flow: open → assigned → resolved / dismissed
- Filterable by site, status, assignee
Audit log + SOC 2 readiness
Append-only log of every state-mutating action across the dashboard and API: API key rotations, member changes, policy edits, integration deliveries, exception transitions. Built for the SOC 2 Type II audit that's currently in progress.
- Action / resource / actor / IP / user_agent captured
- Searchable, filterable viewer
- Service-role inserts only
- Customer members read-only via RLS
Integrations framework
Push verification events into claims platforms. Adapter registry pattern with retry queue (1m / 5m / 30m / terminal). Guidewire ClaimCenter adapter stub shipped; Duck Creek and Mitchell on roadmap.
- verify_ai_integrations + verify_ai_integration_deliveries
- Credentials in Supabase Vault
- Customizable field mappings
- Same retry pattern as outbound webhooks
BI export
Signed-URL CSV or JSON exports of verifications, exceptions, or audit log for downstream BI. 50,000-row cap per request; respects site-scoping. v2: direct connectors via Fivetran / Stitch.
- 3 scopes: verifications, exceptions, audit_log
- since / until date filters
- Signed URL with 1-hour TTL
- Roadmap: NDJSON + scheduled exports
Vertical integrations
Channels and adapters built for specific verticals: WhatsApp for EMEA/LATAM insurance, BOL OCR for freight, chargeback defense for retail suppliers, TMS adapters for trucking.
WhatsApp capture
PreviewConversational photo capture over Twilio WhatsApp Business API. State-machine guides the user through each shot (greeting → consent → instructions → captures → review → submitted). Critical for markets where WhatsApp is the default channel.
- HMAC-SHA1 signature verification
- Twilio media download + processing
- 24h idle → abandoned
- Consent + opt-out keyword handling
BOL OCR
Extract bill-of-lading fields from a photo: BOL number, PRO, weight, freight class, addresses, pieces. Powers chargeback-defense automation. Gemini structured-output with Zod-validated response.
- BOL number + PRO extraction
- Weight, freight class, pieces
- Ship-from + ship-to addresses
- Confidence score per extraction
Chargeback defense
Auto-draft dispute letters for Walmart OTIF, Amazon Vendor, Target, Kroger deductions. 10 prebuilt templates seeded; customer-specific templates take precedence. Dispute drafter joins verification → shipment → template.
- 10 seeded templates (Walmart 22/24/25/90/MABD, Amazon 5/8/11, Target, Kroger)
- Mustache-style placeholders
- Customer overrides for retailer-specific language
- SupplyPike outcome webhook (preview)
TMS adapters (McLeod + SupplyPike)
PreviewAdapter stubs ship with the documented config shape for McLeod LoadMaster (REST/IFS) and SupplyPike (Bearer + HMAC inbound webhook). Wire them to your tenant credentials when ready. Cargowise + MyEZclaim on roadmap.
- Stubbed runtime — throws until configured
- Documented IntegrationAdapter interface
- Per-tenant config in verify_ai_integrations
- Same retry framework as Cluster 3
ML platform & streams
Production-grade ML platform: targeted model dispatch, canary rollouts, drift detection, and drive-through camera streams. Built so the team can ship models with confidence.
Model dispatch + canary tiers
Ship different models per asset type, region, customer, or app version. Production / canary / shadow tiers with sticky bucketing via SHA-256 hash of customer_id + device_id. Roll out canaries to 10% of traffic, monitor, promote.
- Targeting JSON: asset_types, regions, app_version, customer_ids
- Canary weights with sticky hashing
- Shadow tier: served only to drift evaluator
- ETag-cached SDK download
PSI drift detection
Population Stability Index over feature + result distributions, computed hourly against a reference snapshot taken at promotion. Sentry alert on PSI > 0.25. Cleanup of recent_inferences runs in the same cron tick.
- Hourly /api/cron/ml-drift
- Reference snapshot at model promotion
- <0.1 stable / 0.1–0.25 minor / >0.25 alert
- 7-day TTL on recent_inferences
Streams API (drive-through)
PreviewPer-frame verification endpoint for fixed drive-through cameras. QR-pair flow mints a long-lived stream_token (bcrypt-hashed in DB) from a short-lived one-time code. Sync verification result returned in <800ms.
- X-Stream-Token header auth
- 10-minute pairing-code TTL, single-use
- Per-stream policy_id binding
- Multi-camera correlation on roadmap
Custom model training
Train your own RT-DETRv2-derived detector on a custom policy. Datasets in verify_ai_ml_samples (COCO JSONB), training via the ml/ Python pipeline, bundle registration via register_bundle.py. Lineage tracked at the bundle level (git_sha, training_dataset_id, gold_eval_results).
- RT-DETRv2-S base architecture
- CoreML + TFLite + ONNX multi-format export
- Per-bundle lineage tracked
- Most customers don't need this — start with shared models
Ready to See It in Action?
Start with a free sandbox account. Test every feature with no credit card required.