MULTIMODAL TRAINING AND EVALUATION DATA. BUILT IN THE EEA
Benchmarks are recorded in quiet rooms. Your users are not in one.
Models that pass the benchmark and fail in production have a data problem.
Condition
Move across the frame to play through the recording
THE CONDITION SPACE Device · Environment · People · Edge case
Four axes decide whether your data survives deployment
You name the range your deployment has to hold. The programme is built to span it, and every sample lands with its coverage recorded.
AXIS 01
Device
Capture across heterogeneous hardware, because a model trained on one tier degrades on another.
- Smartphone tier
- Webcam tier
- USB headset
- In-vehicle microphone
- Clinical and specialist equipment
AXIS 02
Environment
The room, the vehicle, the street. Acoustics, light and interference are specified, not hoped for.
- Clinical
- Automotive cabin
- Industrial, hearing protection
- Contact centre
- Home
- Outdoor and field
AXIS 03
People
Stratified participant pools, so the long tail is represented rather than averaged away.
- Age range
- Gender identity
- Skin tone
- Accent and dialect
- 150+ languages and their regional variants
AXIS 04
Edge case
Targeted capture of the cases a benchmark removes and production does not.
- Rare accents
- Occluded objects
- Low-light scenes
- Ambiguous utterances
- Code-switching mid-sentence
Coverage is agreed before capture begins and delivered as part of the record. Where a required range cannot be covered, YPAI says so during scoping rather than after delivery.
COVERAGE, RECORDED Device × Environment · 30 conditions
A benchmark covers one corner. A programme covers the grid
Every cell is a named condition. Public benchmarks cluster in the quiet corner. Your programme is specified against the cells your deployment will actually meet, and ships the proof in the record.
A LIVE TEST IN YOUR BROWSER The quiet room · The world
Hear it in your own voice
Record three seconds in your quiet room. The same take is transcribed clean, and again with the world mixed in at a chosen level. The words that survive are the difference training data has to cover.
The quiet room reference
start route guidance to the harbour terminal
The world Automotive cabin · motorway
stop root guidance to harbour tunnel
Runs in this tab, on your device Example shown until you run it. Illustrative, no customer data.
WHAT YPAI COLLECTS Audio · Video · Image · Text · Sensor · Physical AI
Start from the data your model needs
Six channels, one operation. Each line names what is captured and where its depth lives, and no modality ships without the same record behind it.
-
Speech and audio
Read and spontaneous speech at 48 kHz / 24-bit, in-cabin and far-field, code-switching, wake words, professional voice and TTS, with transcripts verified by native reviewers.
Speech data -
Video
First-party video for tasks that depend on motion and sequence: driver monitoring, gesture and body pose, multi-camera synchronisation with documented lighting and occlusion.
Video data -
Image
Still-frame capture across lighting, viewpoint, device and demographic variation, collected to your specification with consent recorded per contributor.
Image data -
Text and document
Documents, parallel corpora and written interaction data across 150+ languages, produced by native speakers under documented rights.
Scope this capture -
LiDAR, 3D and sensor
LiDAR, radar, IMU, thermal and IoT streams with calibration-verified alignment and time-synced capture for ADAS, robotics and spatial AI.
Image, 3D and sensor data -
Physical AI and robotics
Demonstration and episode data for robots and embodied systems, indoor and outdoor, captured with stereo camera and RTK GPS rigs.
Physical AI data
- Multimodal collection
- Low-resource languages
- Conversational and multi-speaker audio
- Voice talent and TTS
- Automotive in-cabin
- Egocentric and robot demonstration video
- Human motion and gesture
- Biometric and face collection
- Clinical and healthcare
- Geospatial
- Field and in-person collection
- Survey collection
- Expert-written text and prompts
- Synthetic and augmented data
- Ready-made datasets
Working with a modality not listed here? YPAI designs capture protocols across data types; describe the deployment and the constraint it runs under.
Scope a programmeHOW A PROGRAMME RUNS Coverage specification · Capture · Quality gates · Delivery
Every sample arrives with its record
One programme, four steps. Each step stamps the record that travels with the data, so every delivery arrives ready to inspect.
One sample. Smartphone tier · Outdoor and field · Low-light scenes
-
Coverage specification
The four axes are turned into quotas: who, where, on what device, and how much of the edge.
- Quotas Per axis, agreed before capture
- Acceptance Criteria written before work starts
-
Capture
Identity-verified contributors record under the specified conditions, consent per record.
- Consent GDPR Article 7, per record
- Provenance SHA-256 manifest per sample
-
Quality gates
Automated checks plus human review against the agreed sampling and acceptance plan.
- Review Native reviewer on sampled work
- Verdict Accepted, or reworked until it passes
-
Delivery
Raw and processed data with annotation layers, in the formats and location you named.
- Formats Raw, processed, annotation layers
- Destination S3, GCS, Azure or on-premise
The architecture, formats, controls and delivery boundaries are described in full. Read the technical brief →
WHO RECORDS Recruited · Identity-verified · Qualified · Calibrated · Reviewed
Qualified contributors, not a crowd
Each contributor is recruited, identity-verified and screened for the specific programme they record in.
-
Recruited
Sourced for the programme's languages, demographics and conditions
-
Identity-verified
Verified identity on file, consent per record
-
Qualified
Passes the programme's own screening tasks before production
-
Calibrated
Agreement tracked against reference work during production
-
Reviewed
Independent native review on sampled work
Programmes draw on a network of 210,000+ registered contributors; each programme uses only the contributors qualified for it.
CONSENT, PROVENANCE AND RESIDENCY Legal basis · Consent · Provenance · Residency · Retention · Agreements
The record your legal and security review will ask for
Governance artefacts are produced by the operation as it runs. Depth and format follow your risk profile and are agreed during scoping.
What your review will ask, and what is on file
-
01 What is the legal basis for each sample?
Legal basis Consent, legitimate interest or contractual necessity, documented per sample
-
02 Can you show consent, per contributor?
Consent GDPR Article 7, collected on the YPAI platform, withdrawable
-
03 Where did each sample come from?
Provenance Per-sample manifest, aligned to EU AI Act Article 10 data governance
-
04 Where does the data reside, and under whose jurisdiction?
Residency EEA by default, Norwegian jurisdiction; US or on-premise by arrangement
-
05 What happens to the data when the engagement ends?
Retention Retention, withdrawal, erasure and closeout defined in the engagement
-
06 What governs the engagement on paper?
Agreements DPA signed, SCCs available for cross-border transfer
Every decision stays bound to the frame.
- captured consent recorded · 48 kHz · 24-bit · far-field Proprietary video and audio collection platforms with verified per-contributor consent
- labelled 3 objects · 1 region · adjudicated against the guideline Native human-review console as the platform core
- inspected sampled lot · coverage gap · 3 items remediated Self-hosted European annotation servers
- evaluated response B accepted on the rubric Credentialed domain reviewers where the work requires them
- delivered manifest · integrity check · erasure within 30 days Residency, subprocessors and transfers defined per engagement
Every delivery carries the records the engagement requires and an integrity check on what is delivered.
INDUSTRIES WE SERVE Healthcare · Automotive · Education · Robotics
Every industry, four with dedicated depth
Every programme inherits the same capture discipline, whatever the industry. Four run deep enough to carry their own collection pages; the rest start at the intake below.
All industry solutions →-
Healthcare
Speech and imaging pipelines for clinical workflows, processed in the EEA by default. Radiologist-vetted annotation, consent-verified participant pools.
EEA processing by default, per-project residency controls
Healthcare -
Automotive
In-cabin voice, multilingual driver interaction, sensor-fused datasets for ADAS validation under cross-environment capture.
Multi-device, cross-condition coverage
Automotive -
Education
Multilingual learning content, accent coverage, accessibility datasets for K-12 and higher-ed applications.
Nordic and European language breadth
Education -
Robotics and industrial vision
Defect detection, robotic-arm telemetry, manufacturing-floor edge cases. Sensor and vision multi-modal pipelines.
Cross-sensor labelling on self-hosted CVAT
Robotics and industrial vision
PILOT TO PRODUCTION Collection · Annotation · Validation · Human and model evaluation · Accepted
One pilot against your requirement.
A pilot is scoped to your specification, quality thresholds and acceptance criteria, and reviewed against them before anything scales. Scope and commercial terms are agreed before it starts. Production is a separate decision, taken after the pilot review.
The pilot fixes
- Scope of work
- Technical requirements
- Acceptance criteria
- Data protection and rights
- Commercial structure
- Remediation and change control
- scope your specification and acceptance criteria
- terms scope and commercial terms agreed before it starts
- review against the agreed criteria, in a pilot workspace
- production a separate decision, taken after the review
Production: a separate decision, taken after the pilot review
Scope a pilotENGAGEMENT PROCESS
From scoping to production dataset
- 01 Describe your use case What modalities, environments, and constraints define your deployment?
- 02 Technical assessment We evaluate feasibility, define QA rubrics, and identify governance requirements.
- 03 Pilot delivery Small-scale data delivery to validate quality gates, formats, and integration.
- 04 Production scale Full dataset delivery with ongoing QA, versioning, and support.
GDPR Article 7 · EU AI Act Article 10 · DPA included
GOVERNANCE
Consent, provenance, and audit readiness
What governance artefacts can be delivered with a dataset?
We can deliver documentation aligned to your risk profile: consent records, provenance logs, demographic breakdowns, QA audit trails, and data processing agreements. Format and depth depend on your compliance requirements.
Can data be collected under a specific legal basis?
Yes. We support consent-based collection, legitimate interest frameworks, and contractual necessity depending on jurisdiction and use case. Legal basis is documented per-sample.
What data residency options are available?
Primary operations are EU-based (Norway). We can arrange US residency or on-premise delivery for restricted deployments. Residency requirements are defined in the project scope.
How is participant consent managed?
Consent is collected through our platform with clear disclosure of data use, retention, and rights. Participants can withdraw, and we support downstream anonymisation or deletion requirements.
Can YPAI sign a DPA or work under our existing agreements?
Yes. We routinely sign DPAs and can operate under client-provided agreements where feasible. Standard Contractual Clauses (SCCs) are available for cross-border transfers.
We will define the data product required for your deployment context and constraints.
If YPAI is not the right fit, we will say so directly.