Rivorix AI

Convert real industrial workflows into structured training data for robots and vision-language models.

End-to-end workflow: Capture → Evaluate → Deploy.

How Rivorix Works

A streamlined pipeline transforming raw footage into high-quality training datasets.

1. Upload Footage

Drag-and-drop video uploads.
Secure, fast, and reliable.

2. Evaluation

Multi-phase pipeline checks for quality, safety, and robotics readiness.

3. Admin Insights

Gain oversight with comprehensive dashboards. Filter by quality tiers and project status.

Organize Your Data Collections

Create projects, invite team members, and manage permissions.
Rivorix scales with your organization.

Plastic Line

Plastic manufacturing training dataset 1.

4 Members

1,240 Videos

Construction

Battersea Powerstation Phase E.

8 Members

856 Videos

Create New Project

Start a new data collection effort

5-Phase Evaluation Pipeline

Every uploaded video goes through a rigorous assessment.

We filter out noise so your engineers get only the best training data.

Pipeline Metrics

Avg Evaluation Time~45 sec

Accuracy95%

Phase 1: Gatekeeper

Pass / Fail (Auto-Reject)

Catch bad footage before spending GPU/LLM time.

Resolution ≥720p

FPS ≥24

No PII/Unsafe actions

OCR Legibility

Phase 2: Quality Tier Scoring

Tier 1-5 Score

Scoring 1–5 mapped to Gold, Silver, Bronze, Marginal, or Reject.

Narration Quality

Process Completeness

Object Engagement

Field of View

Phase 3: Robotics Readiness

Pass / Fail

VLM Verdict: Can a robot learn from this clip?

Action Completeness

Camera Stability

Contextual Description

Obstruction Check

Phase 4: Interaction Check

Valid / Invalid

Validating HMI/Touchscreen interactions with IoU checks.

Button Presence

Bounding-box Validity

Finger Occlusion Ratio

IoU Confidence

Phase 5: Final Classification

Ideal / Borderline / Reject

Consolidated verdict for dataset inclusion.

First-person POV

Safety Compliance

Workflow Clarity

Real-world Objects

Admin Dashboard

Overview of all video evaluations

Total

1,248

Ideal

856

Borderline

312

Reject

Evaluation Log

Quality View

Video Name	Project	Classification	Rating	Date
training_demo_01.mp4	Rivorix Core	ideal	4.8	Oct 24
hand_assembly_v2.mp4	Mfg Pilot	borderline	3.2	Oct 23
safety_test_fail.mp4	Safety Ops	reject	1.5	Oct 22
robot_arm_calib.mp4	Rivorix Core	ideal	4.9	Oct 21

Global Data Oversight

Rivorix gives your ops team a centralized command center. Monitor dataset health, catch systematic quality issues, and export training-ready manifests in one click.

Unified Evaluation Log

Search and filter across every project to find specific edge cases or failure modes.

Quality Classification

Instantly spot Ideal vs Rejected clips based on your custom rubric.

Quality Promise

Trusted, robotics-grade data backed by layered governance.

Automated Gemini scoring + human QA + detailed telemetry keep your datasets safe, explainable, and repeatable.