Software Engineering & Digital Products for Global Enterprises since 2006
CMMi Level 3SOC 2ISO 27001
View all services
Staff Augmentation
Embed senior engineers in your team within weeks.
Dedicated Teams
A ring-fenced squad with PM, leads, and engineers.
Build-Operate-Transfer
We hire, run, and transfer the team to you.
Contract-to-Hire
Try the talent. Convert when you're ready.
ForceHQ
Skill testing, interviews and ranking — powered by AI.
RoboRingo
Build, deploy and monitor voice agents without code.
MailGovern
Policy, retention and compliance for enterprise email.
Vishing
Test and train staff against AI-driven voice attacks.
CyberForceHQ
Continuous, adaptive security training for every team.
IDS Load Balancer
Built for Multi Instance InDesign Server, to distribute jobs.
AutoVAPT.ai
AI agent for continuous, automated vulnerability and penetration testing.
Salesforce + InDesign Connector
Bridge Salesforce data into InDesign to design print catalogues at scale.
HumanDISC
AI-powered behavioral assessments and DISC profiling for smarter hiring.
View all solutions
Banking, Financial Services & Insurance
Cloud, digital and legacy modernisation across financial entities.
Healthcare
Clinical platforms, patient engagement, and connected medical devices.
Pharma & Life Sciences
Trial systems, regulatory data, and field-force enablement.
Professional Services & Education
Workflow automation, learning platforms, and consulting tooling.
Media & Entertainment
AI video processing, OTT platforms, and content workflows.
Technology & SaaS
Product engineering, integrations, and scale for tech companies.
Retail & eCommerce
Shopify, print catalogues, web-to-print, and order automation.
View all industries
Blog
Engineering notes, opinions, and field reports.
Case Studies
How clients shipped — outcomes, stack, lessons.
White Papers
Deep-dives on AI, talent models, and platforms.
View all resources
About Us
Who we are, our story, and what drives us.
Co-Innovation
How we partner to build new products together.
Careers
Open roles and what it's like to work here.
News
Press, announcements, and industry updates.
Leadership
The people steering MetaDesign.
Locations
Gurugram, Brisbane, Detroit and beyond.
Contact Us
Talk to sales, hiring, or partnerships.
Request TalentStart a Project
AI & Machine LearningHR Tech & Recruitment

Dense keypoint annotation and gaze tracking dataset to power ForceHQ's anti-cheating computer vision model.

We orchestrated an advanced AI Data Labeling pipeline for dense facial keypoints, enabling a custom computer vision model to track subtle gaze vectors and head pose for remote assessment proctoring.

OpenCV · MediaPipe · Python
Client: ForceHQ
Facial Landmark Data Labeling for AI Proctoring

Project Overview

ForceHQ needed an automated AI proctoring system capable of detecting suspicious behaviors during remote assessments—specifically, whether a candidate is looking at a secondary monitor, reading off-screen notes, or receiving unauthorized help. After off-the-shelf computer vision models suffered from high false-positive rates due to unpredictable remote webcam environments (poor lighting, varied camera angles), ForceHQ realized they needed a custom model. To build a highly accurate model that could track subtle eyeball movements, neck rotation, and micro-expressions, they first required a meticulously labeled, bias-free dataset of facial landmarks.

Dense Facial Keypoint Annotation

Our AI Data Labeling team processed thousands of hours of diverse webcam footage, mapping complex 68+ point facial landmarks to track eyes, nose, and jawline with pixel-perfect accuracy.

Gaze & Head Pose Labeling

We went beyond standard landmarks by annotating precise pupil vectors and calculating head pose orientations (pitch, yaw, roll) to establish baseline attention metrics.

Have a similar challenge?

Our experts can help you build custom integrations and plugins tailored to your business workflows.

Book a free consultation

Dataset Balancing & Bias Mitigation

Implemented rigorous QA workflows to ensure the dataset was heavily balanced across different skin tones, facial hair, glasses, and lighting conditions to completely eliminate algorithmic bias.

Custom Computer Vision Training

Using this high-fidelity labeled data, we successfully trained a custom computer vision model that accurately flags suspicious assessment behavior in real time while drastically reducing false positives.

Key Challenges

01

Challenge 1

Off-the-shelf models failed under the diverse and unpredictable conditions of remote webcams (low resolution, poor lighting, varied angles).

02

Challenge 2

Tracking subtle eyeball movements and neck rotation (pitch, yaw, roll) requires pixel-perfect, dense keypoint annotation.

03

Challenge 3

Algorithmic bias was a major risk; the dataset needed to perform equally well across diverse skin tones, facial features, and the presence of glasses.

04

Challenge 4

High false-positive rates in automated proctoring severely degrade the candidate experience, demanding exceptional model precision.

Results & Outcomes

1M+
Facial Keypoints Annotated
85%
False Positives Reduced
100%
Demographic Balancing
<50ms
Gaze Detection Latency
FAQ

Frequently Asked Questions

Common questions about this topic, answered by our engineering team.
Standard object detection (bounding boxes) cannot determine where a person is looking. Dense keypoint annotation maps the exact geometry of the face, allowing models to calculate the precise angle of the neck and the direction of the pupils, which is essential for detecting off-screen gazing.
We annotate specific anchoring landmarks (like the tip of the nose, corners of the eyes, and chin). By comparing the 2D coordinates of these points across frames against a standard 3D facial model, we can mathematically calculate the pitch (nodding), yaw (shaking), and roll (tilting) of the head.
We implemented a strict stratified sampling process during data collection and labeling. The QA team audited the dataset to ensure equal representation across the Fitzpatrick scale (skin tones), genders, ages, and edge cases like thick glasses, facial hair, and extreme low-light environments.
We utilized advanced computer vision annotation tools like CVAT and specialized plugins that allow annotators to project a base mesh onto a face and manually adjust the individual nodes. This drastically speeds up the process compared to plotting 68 individual points from scratch on every frame.
Because the model was trained on high-fidelity, diverse data rather than synthetic or limited datasets, it learned to differentiate between a candidate thinking (looking up briefly) versus a candidate reading from a secondary monitor, resulting in an 85% reduction in false cheating flags.
Have a similar challenge?

Let's build your success story.

A 30-minute call with a principal engineer. We'll discuss your challenges, propose architecture, and outline a roadmap.

Talk to a strategist
Have a similar project? Let's discuss your requirements.
Book a call
EmailWhatsApp