Industrial AI Annotation

Precision Labels.
Recursive
Intelligence.

Production-grade training data for industrial vision AI — combining foundation model pre-annotation, expert human refinement, and a Taiwan-anchored trust layer built for global compliance.

Request a Pilot View Our Work

LIVE ANNOTATION PIPELINE

Images

99.2%

Accuracy

Defect Types

Defect Detection Solder Bridge Annotation Open Circuit Detection Missing Hole Labeling Mouse Bite Classification Instance Segmentation Bounding Box Annotation Keypoint Labeling 3D Point Cloud Video Tracking Defect Detection Solder Bridge Annotation Open Circuit Detection Missing Hole Labeling Mouse Bite Classification Instance Segmentation Bounding Box Annotation Keypoint Labeling 3D Point Cloud Video Tracking

What We Do

Annotation services engineered
for industrial precision

From defect detection to dimensional measurement — we transform raw visual data into high-fidelity labels that power the next generation of industrial vision systems.

Defect Detection Annotation

Pixel-accurate labeling for surface defects, cracks, voids, scratches, and contamination across PCB, semiconductor, and manufacturing inspection pipelines.

Instance Segmentation

Precise polygon and mask annotations for complex, overlapping objects — enabling models to distinguish individual components on dense assembly lines.

Bounding Box & Classification

High-throughput object detection labeling with multi-class taxonomies, optimized for real-time quality inspection and sorting systems.

Keypoint & Measurement

Landmark and dimensional annotation for alignment verification, assembly validation, and robotic pick-and-place calibration.

3D Point Cloud Labeling

Volumetric annotation for LiDAR and depth sensor data used in warehouse automation, robotic navigation, and industrial environment mapping.

Video & Temporal Tracking

Frame-by-frame object tracking and event annotation for motion analysis, production line monitoring, and predictive maintenance vision systems.

Portfolio

Real annotations.
Real ground truth.

Over 20,927 PCB images annotated across 6 defect categories — built as proprietary ground truth for our self-evolving detection pipeline.

Bounding Box · Classification

Missing Hole Detection

Locating absent drill holes in PCB manufacturing — critical for board functionality and electrical connectivity validation.

Polygon Segmentation

Short Circuit Annotation

Pixel-level segmentation of unintended conductive bridges — the most common and costly PCB defect in high-density assemblies.

Multi-class · 20,927 Images

Full Defect Dataset

6-class ground truth dataset covering Mouse Bite, Open Circuit, Short, Spur, Spurious Copper, and Missing Hole — built for YOLO fine-tuning.

How It Works

The recursive refinement loop

Every dataset passes through our AI + Human iterative pipeline — each cycle sharpening accuracy, consistency, and domain alignment.

Ingest & Scope

We analyze your raw data, define label schemas, and establish acceptance criteria.

AI Pre-Annotation

SAM2, YOLO, and GroundingDINO generate initial labels at machine speed.

Human Refinement

Trained domain annotators correct, refine, and validate every label against your quality spec.

QA & Delivery

Automated consistency checks, cross-validation, and export in COCO / YOLO / custom format.

Industries

Built for sectors where accuracy is non-negotiable

We serve clients across the industrial spectrum — from high-mix electronics to heavy manufacturing — where mislabeled data means real production loss.

PCB & SMT

Solder joint, component placement, bridge / tombstone detection

Semiconductor

Wafer defect mapping, die-level classification, packaging inspection

Automotive

Paint finish, weld seam, assembly verification across production lines

Metal & Machining

Surface roughness annotation, tool wear, dimensional compliance

Pharma & Medical

Packaging integrity, vial inspection, label verification

Energy & Infrastructure

Corrosion mapping, crack detection on turbines and structural assets

Open-source AI,
industrial expertise

We build on the best open-source foundation models — then fine-tune them on your domain data to achieve accuracy no general model can match.

SAM 2

YOLOv11

GroundingDINO

Qwen3-VL

CVAT

COCO Format

GCP / Taiwan

ISO Pipeline

Why Recursia

Three pillars of our unfair advantage

Taiwan Trust Layer

Your data enters through a Taiwan-based entity with GCP infrastructure in compliant regions. We serve as the legal and operational firewall meeting EU, US, JP, and KR data residency requirements.

AI-First Architecture

14 AI digital workers orchestrate every project — from intake triage to quality scoring. Foundation models pre-annotate at machine speed; domain experts refine. Result: 3× throughput at half the error rate.

III

Recursive Quality Loop

Every annotation cycle feeds back into model fine-tuning and annotator calibration. Accuracy compounds — your tenth batch is measurably better than your first.

Ready to upgrade
your training data?

Start with a free pilot. We'll annotate a sample of your data so you can benchmark quality before committing to a full project.

Precision Labels.RecursiveIntelligence.