Skip to content
Langhui
中文 Book a consultation
Solutions Capture hardware Catalog Quality Insights About Book a consultation

Langhui

SOLUTIONS

A closed loop for training-data delivery

Langhui supports model teams, healthcare organizations, and industry groups with three capabilities: data you can evaluate, collection and expert review you can control, and capture hardware for embodied scale. From task definition through device choice, production, and versioned delivery, the work is built to be accepted.

Book a solution review Choose capture hardware Browse the catalog
ARCHITECTURE

Solution architecture

Buyers and research teams usually start with three questions: whether the training data can be obtained and explained; whether production and final review stay under control; and whether capture hardware and data engineering can ship together for embodied work. The first two layers are below. Capture hardware has its own section.

01 · Data delivery

Four data assets you can evaluate

Medical data Built on licensed hospital partnerships, covering imaging, clinical documents, and structured fields. Before delivery we assess source, license, modality, specialty, and task fit. View details AI Coding Benchmark Training and evaluation data for coding models and agents: task definition, execution environment, checks, and scoring, for post-training and benchmark alignment. View details Legal and judicial data Legal documents, judgments, statute structure, contract review, and professional Q&A, with a documented license boundary. View details Embodied intelligence data Egocentric behavior, humanoid imitation, manipulation, and multi-scene interaction. It can be paired with the TOBI matrix in the capture section. After a pilot, production and versioned delivery follow the agreement. View details
02 · Platform and tools

Production, quality control, and experts

Annotation capacity and supply
Schedulable annotation and QC resources, an in-house platform, and a supplier system, with first pass, review, and final review set by layer.
Medical imaging annotation platform
The base of the medical loop. CT / MR / PET · DICOM by default, deployed on-premises or as a combined capture-and-label training path.
Expert evaluation system
Final review, dispute decisions, and professional evaluation run through the Langhui expert system: onboarding, task work, and settlement.
Capture hardware (TOBI)
The hardware layer for embodied collection at scale. Choose an Ego headset, hand capture, a teleoperation base, and an end effector by task. See the section below and the on-site capture-hardware center.
CAPTURE HARDWARE · TOBI

Capture hardware: the base for embodied collection and labeling

When the delivery is egocentric behavior, hand manipulation, or teleoperation demonstrations, the device sets sync accuracy, scale cost, and reproducibility. The capture layer uses the TOBI matrix: choose an Ego headset, data gloves, a teleoperation base, and an end effector, then connect them to embodied data engineering, QC, and versioned delivery.

How to choose

  • Task before model. First decide whether you need on-device SLAM, millimeter fingertip space, force-feedback teleoperation, or long multi-site deployments, then open that SKU family.
  • Specs follow writing. This page guides by task family. The full catalog, specification sheets, and ordering notes are in the capture-hardware center. Supply follows the latest written confirmation. Specification pages are published in Chinese.
  • Collection and labeling can be one project. Devices, a pilot, labeling, and versioned delivery can share one project, or you can buy only the device matrix or only a data pack.
01 · Ego vision

Head-worn and near-field view

Vision capture for multi-site scale, on-device trajectories, or a project evaluation. A wrist camera can sync with the main unit for fine hand work.

02 · Hand capture

Fingertip space and long sessions

If you need millimeter fingertip coordinates, use magnetic-inertial fusion. For long wireless gestures, use the inertial glove. Metal environments and two hands close together need a check first.

How this joins the embodied data program

Devices answer stable capture, accurate sync, and affordable scale. The embodied data program answers clear labels, checks that pass, and a delivery you can accept. Typical path: task definition, device and scene pilot, labeling and QC, then versioned delivery. You can also buy only devices or only a data pack.

Embodied page · capture hardware Embodied data program Open the full device matrix
EXPERT EVALUATION

Expert evaluation and the business system

Expert work uses the existing business system. The website does not open a parallel application. Individual experts and supplier organizations register, pass review, collaborate, and settle in that system, on the same process as delivery.

Role of the system

Expert review is the gate in the quality loop: guideline confirmation, dispute samples, the accepted final set, and professional evaluation, within the project. Onboarding, tasks, and settlement stay in the business system so responsibility and process can be traced.

  • Individual experts: credential review, domain match, quality review, and payment
  • Supplier organizations: company credit, team management, delivery, and reconciliation
  • Project owners: bring in final review and evaluation by task, without a second process
01

Expert network

Onboarding conditions, review, and the boundary of collaboration, before you enter the business system.

Open the expert network
02

Expert workspace

Registration and login for certified experts. Credentials, tasks, review, and settlement stay on the expert platform.

Sign in to the expert platform
03

Supplier workspace

Registration and login for data-service firms and annotation teams. Company credentials, delivery, and settlement stay on the supplier platform.

Sign in to the supplier platform
MEDICAL DATA

How medical data is assessed

Medical delivery centers on controlled data from partner hospitals. Across the catalog, review these six dimensions.

  1. Source and license. Hospital partnership or sublicense, de-identification rules, and what can and cannot be provided.
  2. Imaging modality. CT / MR / PET by default. Other modalities are assessed per project.
  3. Documents and structure. Record fields, diagnostic reports, checkup structure, and the boundary of knowledge content.
  4. Specialty and findings. Anatomy, disease labels, and whether a reference standard is available.
  5. Task form. Classification, detection, segmentation, report generation, text extraction, and other training goals.
  6. Delivery form. After a controlled sample, agree a hard drive, cloud drive, API, or a project delivery.
ENGAGEMENT

How a project starts

  • Evaluate existing data. Match a catalog direction, then request a controlled sample.
  • Medical platform or training set. In the consultation, say whether you need on-premises deployment or delivery to a model team.
  • Capture hardware and an embodied pilot. Choose in the capture section, or open the capture-hardware center. It can be assessed together with an embodied data pack.
  • Expert review and onboarding. Enter the expert or supplier system through the expert network.
  • The need is still open. Send the task, the data you already have, and the delivery you want. The project team will suggest an entry point.