Real-world data. Structured for AI.
AGTRIX is building a data collection and annotation service for teams working on computer vision, robotics, and embodied AI. We aim to turn real-world activities into structured, documented, quality-checked datasets through carefully defined collection and annotation workflows.
Faster Robots × Higher Accuracy × Smarter Work
Robots learn from examples. Examples need structure.
Video of people doing real tasks only becomes useful training data once it has clear scope, agreed labels, timestamps, documented permissions and a way to check the work. AGTRIX is designed around that gap, starting with small pilots where requirements are validated before anything is scaled.
Task-specific data
Footage and labels scoped to the activity your system needs to learn.
Machine-readable output
CSV and JSON against an agreed schema, with documented label definitions.
Pilot first
Validate task, format and acceptance criteria before committing to volume.
Permissions in view
Consent, provenance and commercial rights are part of scope, not an afterthought.
Services
Services we are designed to offer. Scope, formats, timelines and price are agreed per project.
First-person video collection
Authorized everyday and work tasks such as picking, packing, sorting, scanning or simple assembly. Each project needs a defined scope, approved recording conditions and participant consent. Smartphones may suit early pilots; production cameras follow your specification.
Task categories we plan to offer. None are available yet.
- Kitchen work Coming soon
- Garage work Coming soon
- Workshop and repair Coming soon
- Warehouse packing Coming soon
- Household chores Coming soon
- Cleaning and maintenance Coming soon
- Retail and stockroom Coming soon
- Laboratory bench work Coming soon
- Gardening and outdoor work Coming soon
Need a different task? Discuss a custom pilot.
Video annotation
Start and end timestamps, action labels, object references, task stages, outcomes and exceptions. Object tracking or bounding boxes are available for discussion where specified and feasible. Not every annotation type is included by default.
Dataset preparation
CSV or JSON exports, agreed schema and label definitions, manifests, file naming, annotation guidelines, documentation and versioning where required.
Quality assurance
Required-field, timestamp and schema checks, label consistency review, human review of correctness and a documented issue and correction log.
Custom data pilots
Small scoped engagements to validate task requirements, recording conditions, annotation definitions, format, acceptance criteria and delivery. No price or turnaround is promised before scoping.
Demonstration dataset
DEMONSTRATION SAMPLE — Illustrative data created to demonstrate AGTRIX's proposed workflow. Not a client project or evidence of contracted work.
Test clip. Robotic case-erector workcell, recorded handheld. Draft labels by AGTRIX, pending human review. The person who appears briefly is blurred and the audio is removed. The source and site permission must still be confirmed before showing this to clients. Resolution is low (144×256); a higher-resolution original will look much better.
Head-cam view
Gripper position found by color detection, so it is approximate.
Detail view (zoom)
Magnifies the same footage. A true hand view needs a second synchronized camera recording, which is not part of this clip.
| Start | End | Actor | Action | Object | Outcome |
|---|
Schema
Export preview
If your browser blocks downloads on this page, use the copy buttons. All values are examples only.
Proposed workflow
This is the process we intend to follow. It describes how an engagement would run, not work already performed.
- Requirements discoveryUnderstand the problem and target task. Output: requirements summary.
- Task and dataset specificationDefine footage, labels, formats. Output: draft specification.
- Consent, permissions and collection planningConfirm participants, sites and rights. Output: collection plan and consent forms.
- Small-scale collection or approved footage intakeRecord or receive authorized video. Output: organized source files.
- Annotation guideline creationWrite label definitions and edge-case rules. Output: guideline document.
- Annotation and structured labelingLabel segments to the schema. Output: annotation files.
- Automated validationRun schema and timestamp checks. Output: validation results.
- Human quality reviewReviewers check correctness and flag ambiguity. Output: review notes.
- Correction and final validationFix issues and re-validate. Output: issue and correction log.
- Packaging and deliveryAssemble dataset and documentation. Output: versioned delivery package.
Quality assurance
Acceptance criteria are agreed with each customer before a project begins. We do not publish accuracy rates or guarantees.
Automated checks
- Required fields present
- Valid timestamps; start before end
- Within media duration when known
- Rows match the agreed schema
- Labels belong to the approved set
- Duplicate or conflicting records flagged
Human review
- Action labels match the visible activity
- Object references are appropriate
- Boundaries follow the guidelines
- Outcomes match the evidence
- Ambiguous segments flagged, not guessed
- Corrections documented
DEMONSTRATION DATA — sample report below is illustrative only.
| Check | Illustrative finding | Status |
|---|---|---|
| Required fields | All example rows complete | Pass |
| Timestamp order | Start precedes end in every row | Pass |
| Label set | One label flagged for guideline review | Corrected (example) |
| Human review | One boundary adjusted | Logged (example) |
Deliverables
Final deliverables depend on agreed scope, customer requirements, permissions and applicable rights.
Data
Approved video files where collection and licensing permit; timestamped annotation files; CSV and JSON exports.
Specification
Annotation schema, guideline document and dataset manifest.
Quality
Quality review report and issue and correction log.
Provenance
Consent and provenance documentation suited to the project, plus dataset version information.
Responsible data governance
Principles we intend to apply. We do not claim legal certifications or compliance in every jurisdiction; project-specific requirements should be reviewed with qualified counsel.
Participant consent
- Informed consent where required
- Explain intended and commercial use
- Document permissions
- Honor withdrawal and deletion commitments
Workplace and property
- Permission before recording on private sites
- Respect employer policies
- No covert recording
- Avoid bystanders and sensitive information
Provenance
- Record origin and conditions
- Track versions and documentation
- Separate approved from unusable material
Privacy and security
- Minimize personal data
- Access controls and secure transfer
- Agree retention and deletion
- Redaction where appropriate and feasible
Commercial rights
- Verify rights before delivery or licensing
- Define uses, ownership and terms by contract
- Owning a recording is not the same as permission to license it
About AGTRIX
AGTRIX is an early-stage company being built to support AI, robotics and computer vision teams with structured real-world data. We are developing a practical approach: clearly defined collection requirements, structured video annotations, documented quality checks and carefully considered data provenance. We aim to work with customers to validate their requirements through small, well-scoped pilots before expanding to larger datasets.
Vision: to help make real-world human activities more usable for AI systems by turning video into structured, documented data.
Faster Robots × Higher Accuracy × Smarter Work
That is our tagline and the future we are working toward. We believe robots can become faster and more accurate when they learn from well-structured real-world data, and that this can make work easier and more profitable for the businesses that use them. It is a goal, not a guarantee of results: every pilot starts by agreeing acceptance criteria with the customer.
Target customers include robotics companies, embodied AI teams, computer vision companies, research labs, and industrial and warehouse automation companies. These are categories we want to serve, not existing clients.
Request a pilot
Tell us what you need. Pilot scope, price and schedule are agreed together after we understand the task.