Machine Learning Engineer

About Humyn Labs

Humyn Labs builds the intelligence layer for physical-world AI — systems that perceive, reason, and act in real environments. Our work sits at the intersection of egocentric video understanding, embodied AI, robotics perception, and voice-driven interaction. We move fast, obsess over data quality, and ship at scale.

Humyn Labs converts human action - across sound, sight, movement, and touch - into high-quality multi-modal data signals for physical AI. Operating across 20+ countries in India, southeast Asia, Latin America, and the Middle East: the real-world environments where physical AI deploys, not the labs where it is built.

Our data isn't just collected; it's evaluated, defended, and production-ready. Because before AI can be trusted, its training data must be.

Role Overview

You will own the auto-annotation pipeline that turns raw egocentric footage into training-grade labels. This is the technical core of the product. You'll deploy, fine-tune, and continuously improve a stack of vision models against the hard parts of egocentric data - ego-viewpoint motion, occlusion, gloved hands, motion blur - and keep label quality high at production scale.

What You Will Work On

• Own and improve the perception stack: object detection (YOLO, Grounding DINO), segmentation (SAM / SAM2), hand-pose estimation, and related models.
• Fine-tune and train models on our egocentric data to close domain gaps off-the-shelf SOTA doesn't handle — and own the results, not just the experiments.
• Build face/identity redaction (e.g., SCRFD-based) into the pipeline so every delivery is consent-compliant at scale.
• Stand up evaluation and QA harnesses: measure label quality, track regressions and drift against defined thresholds, and gate releases on them.
• Deploy and run inference at scale on AWS (batch orchestration, throughput tuning, cost control) to hit high-volume daily SLAs.
• Ship to production and stay on the hook for accuracy, latency, and reliability in delivered datasets.

You Must Have

• You've shipped vision models to production — not just notebooks or benchmarks.
• Hands-on experience across the perception stack: YOLO, Grounding DINO, SAM/SAM2, SCRFD, VLMs, or equivalents.
• Real fine-tuning / training experience — you can take a base model and make it work on a new domain, including data, loss, and eval decisions.
• Strong Python + PyTorch; comfortable owning training and inference code.
• Medium-level MLOps: containerization, batch inference pipelines, and orchestration on AWS.

Nice to have

• Egocentric, AR/VR, or robotics data experience.
• Awareness of SLAM/VIO and IMU-synced multimodal data.
• Familiarity with robot-learning formats (LeRobot, RLDS, MCAP).

How we work

• Fast-paced startup.
• High ownership, pragmatic decisions, and a strong bias toward shipping.
• You'll have real scope from day one.