SPORTS TECHNOLOGY · CUSTOM SOFTWARE · COMPUTER VISION

Combining computer vision and expert review for goalkeeper analysis.

GKM Tech prepares uploaded goalkeeper footage with detection, tracking and pose data, then gives specialist analysts the tools to review technique and publish structured scores and feedback.

GKM Tech analyst workspace reviewing goalkeeper tracking and pose evidence frame by frame
GKM TECH · Product MVP
Client
GKM Tech
Sector
Sports performance
Engagement
Product MVP
Status
Human-in-the-loop MVP
Discuss a similar projectExplore Custom software development

PRODUCT

End-to-end

Submission, processing, review and feedback

MEDIA

Asynchronous

Staged video processing outside the request cycle

SCORING

Human-controlled

Specialists publish the score

PROJECT SNAPSHOT

The engagement in three parts.

01

Situation

Goalkeepers wanted structured technical feedback from ordinary training or match video, but interpreting every clip manually made consistent analysis difficult to scale.

02

What we delivered

A customer platform, subscription and submission workflow, asynchronous computer-vision backend, annotation interface and analyst-controlled performance card.

03

Our responsibility

We owned the customer product, processing pipeline, analyst workspace and commercial foundation.

What shipped

A working human-in-the-loop MVP connecting customer video submission to computer-vision preparation, expert review and structured feedback.

  • The product includes both the commercial customer surface and the specialist internal workflow.
  • The boundary between automated evidence and analyst-controlled scoring remains explicit.
Working with TW was an absolute pleasure. We hired them to build the MVP for an AI-powered Computer Vision platform and they exceeded all expectations.

Riccardo C.

GKM Tech

Eyezilla operator feedback

HUMAN-IN-THE-LOOP WORKFLOW

Automation prepares the evidence; the specialist owns the score.

The expert decision is the pivot between computer-vision preparation and customer-facing feedback.

  1. 01 · System stage

    Customer upload

  2. 02 · System stage

    Queued processing

  3. 03 · System stage

    Frames, detections and pose

  4. 04 · Human decision

    Analyst review

  5. 05 · Human decision

    Scoring decision

  6. 06 · Human decision

    Published GKM Card

SELECTED SYSTEM EVIDENCE

Connecting customer submission to specialist judgement.

GKM Tech customer dashboard showing the premium analysis archive and subscription route

01

A customer submission and progress experience

The React application covers authentication, subscription, video submission, processing state, feedback history and the visual GKM Card. Supabase provides managed identity, database and storage foundations, while Stripe and transactional email support the commercial workflow.

Why it mattered

The analysis service became a customer product rather than an internal computer-vision script.

GKM Tech analyst workspace for goalkeeper classification, pose correction and annotation review

02

An annotation and expert-scoring workspace

Analysts can inspect the prepared footage, correct or confirm the relevant player evidence and apply the scoring framework. The published GKM Card reflects specialist review rather than an opaque model output.

Why it mattered

The MVP protects credibility while building the labelled evidence needed for future automation.

ENGINEERING DECISIONS

Automating preparation without automating expertise.

Make the human review boundary explicit

The system does not imply that pose estimation can produce a final performance judgement without context. Automation prepares the evidence and workflow; the analyst owns the published technical score.

Product trust

Separate the customer and analysis applications

The customer-facing product and computer-vision backend can evolve independently. The public application remains responsive while Python services and GPU-compatible tooling handle media and model workloads.

Architecture

Build annotation into the operating product

Review corrections are not discarded as one-off administrative work. The annotation interface creates a route for improving the dataset and evaluating where automation is reliable enough to expand.

Model development

RESPONSIBILITY

Automation and expertise had different owners

Client / domain expert

  • Scoring concept
  • Technique criteria
  • Specialist review

Tested Works

  • Customer product
  • Processing pipeline
  • Analyst workspace

Shared decisions

  • Score representation
  • Automation boundary
  • Acceptance

Working human-in-the-loop MVP

SERVICES USED

Custom software developmentMVP developmentComputer vision systemsProduct design

SELECTED TECHNOLOGY

ReactTypeScriptFastAPIPythonPostgreSQLSupabaseYOLOBoT-SORTRTMPoseStripe

RELEVANT EXPERIENCE

Relevant if your situation includes…

You need to turn specialist expertise into a repeatable software workflow.

You need computer vision to prepare evidence without overstating model judgement.

You need a customer product and an internal operations tool to work as one service.

RELATED WORK

Two adjacent examples.

Person completing a rehabilitation exercise with body pose landmarks overlaid

ALLIMB

Proving camera-guided rehabilitation on everyday devices.

Working mobile PWA with live pose analysis, repetition counting and corrective feedback.

View case study
Fixed-camera tennis court with calibrated geometry, player tracking and active-play retention state

TENNIS VIDEO INTELLIGENCE

Finding active play in fixed-camera tennis recordings.

Court-aware player tracking and explainable active-play timelines.

View case study

START A CONVERSATION

Does your product need AI preparation and expert judgement to work together?

We can design the review boundary, build the operating workflow and create the evidence needed for later automation.

Discuss your productExplore custom software development