Capabilities
How a video dataset gets made, labelled and proven.
Four steps, with the checks and the records each one produces. The five pillars below are the services built on them.
The pipeline
Source, annotate, evaluate, prove.
- STEP 1
Source
Footage from 15+ vetted partners, or custom capture to your specification, with the licensing chain and consent basis documented.
- STEP 2
Annotate
Labels at five levels, from the whole video down to the pixel, all using one taxonomy, manipulation-v2.3.
- STEP 3
Evaluate
Inter-annotator agreement measured on every project, gold-standard clips per domain, and a QA evidence pack.
- STEP 4
Prove
A SHA-256 Merkle manifest, C2PA content credentials and EU AI Act Article 10 documentation, sealed with the delivery.
Why it matters
Three problems that show up once the data is in storage.
Structure
Raw footage has no schema.
Without labels and a taxonomy, terabytes of video have no training value.
Quality
Labels drift when nobody measures them.
Without feedback loops and agreement scores, quality slips a little more with every batch.
Compliance
Regulators now ask where data came from.
The EU AI Act expects documented origin and collection for the training data of high-risk systems.
What we offer
Five pillars, side by side.
Each one has its own capabilities, a named product and an outcome. The homepage shows what each is used for.
- PILLAR 01
Rights, consent, & provenance
VeriPhi PassportCapabilities
- Rights-cleared licensing chain, with consent records
- Hardware-rooted C2PA signatures
- Device fingerprinting & tamper detection
- EU AI Act data documentation
- Sensor-to-cloud audit trail
Outcome
Defensible training data, ready for EU AI Act compliance.
Use cases and detail - PILLAR 02
Data & annotation
VeriPhi DatasetsCapabilities
- Off-the-shelf and custom data collection
- Targeted sourcing from 15+ vetted partners
- Egocentric and spatial-intelligence coverage
- Multi-sensor alignment (video, IMU, depth)
- Vision metadata at shot, scene, segment and full-video level, with indexing
- Action, object and physical-design labels
- Scene graph generation
Outcome
Targeted training data delivered fast, with precise multi-modal alignment.
Use cases and detail - PILLAR 03
Workflow auditor & advisor
VeriPhi JudgeCapabilities
- Automated RLAIF & preference scoring
- Enterprise rules-engine optimization
- Issue detection, deduplication & fixes
- Workflow streamlining
- Expert-in-the-loop rubrics
Outcome
Optimized, compliant workflows with less rework and lower error rates.
Use cases and detail - PILLAR 04
Benchmarks & standards
VeriPhi BenchmarksCapabilities
- Real-time, time-synced benchmarking
- Internal & external safety baselines
- Best-practice ontology standards
- Deep data discovery & indexing
Outcome
Confident model evaluation, safety checks and continuous deployment.
Use cases and detail - PILLAR 05
Tailored SLMs & Custom Engine
VeriPhi EnginesCapabilities
- Domain-specific SLMs & micro-models
- Provenance & policy enforcement models
- Spatial & vision-language-action (VLA) models
- Edge-deployable specialized weights
- Customer-tailored model stacks
Outcome
Task-optimized, low-latency models tailored to specific enterprise needs.
Use cases and detail
Annotation
Five levels of detail, one taxonomy.
The same taxonomy, manipulation-v2.3, is used at every level.
- 01 VideoSessions and datasets, sourced and validated.
- 02 SegmentAction buckets, scene state, your taxonomy.
- 03 FrameCamera craft, IMU sync, depth.
- 04 ObjectBoxes, 21-point hands, grip and force.
- 05 PixelError-level, sensor-noise and entropy checks.
Checked at every level: inter-annotator agreement (target above 0.85), a SHA-256 and C2PA sealed manifest, and EU AI Act Article 10 documentation.
Modalities
The footage we work with.
Working on something that is not listed here? Tell us what you are building. We are exploring new use cases and welcome them.
- EgocentricHead- and wrist-mounted cameras.
- Sensor-synced egocentricVideo paired with IMU accelerometer and gyroscope streams.
- Multi-cameraRigs of 2 to 6 cameras, synchronised to within half a second.
- CinematicBroadcast and cinematic RGB video.
- 3D captureSpatial capture using COLMAP structure-from-motion.
Delivery
Three ways to take it.
As a managed service
We run sourcing, annotation, evaluation and proof for you, and you receive finished batches.
As data only
Take the dataset with its manifest and labels, and run your own training.
In your environment
Deploy the pipeline inside your own infrastructure.
Scope a project
Tell us what you are building.
A couple of lines is plenty. A person on our team reads every message and replies within 48 hours.
