Hardware Engineer

Humyn Labs; Bengaluru; On-site; INR 1,800,000 - 2,400,000; Full time

Job summary

Humyn Labs is building a custom stereo camera system as part of its physical-world AI data platform. You’ll own sensor selection, synchronization, calibration, and geometry-grade data for robust perception models.

Work on embedded hardware like Raspberry Pi and NVIDIA Jetson, design metrology-grade calibration, and ensure field reliability. This role rewards hands-on ownership and solving hard engineering problems in real environments.

Qualifications

• Hands-on experience building multi-camera stereo systems.
• Strong knowledge of camera geometry and calibration.
• Experience with visual-inertial systems (VIO/SLAM).
• Proficiency in C++ and/or Python; ability to optimize performance.

Responsibilities

• Design and build a custom stereo camera end to end.
• Architect synchronized dual-camera capture with hardware triggering.
• Fuse vision with high-rate inertial sensing for VIO/SLAM.
• Own on-device video encoding and bandwidth management.
• Develop metrology-grade calibration with sub-pixel accuracy.
• Take the device from bench to field and validate reliability.

Skills

Embedded systems

PCB bring-up

Sensor interfacing

Stereo camera design

Dual-camera capture

Hardware triggering

Vision-IMU fusion

Video encoding

Calibration pipeline

Sub-pixel accuracy

Field deployment

Camera geometry

VIO/SLAM

Stereo depth estimation

Real-time pipelines

C++ and/or Python

Robotics background

Jetson deployment

Firmware/FPGA

Production readiness

Tools

Raspberry Pi

NVIDIA Jetson Orin

FPGA

PCB bring-up

Job description

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.

About The Role

We build the physical-world data infrastructure that trains the next generation of humanoid robots — which means the quality of everything downstream starts at the sensor. We're building a custom stereo capture device from the ground up, and we're assembling a small, high-ownership pod to own it end to end: the optics and sensors, the synchronized capture stack, the calibration that makes the data trustworthy, and the perception models that turn raw stereo into geometry.

This is not integration work on an off-the-shelf rig. You'll be making the hard calls — which sensors, which optics, how to sync, how to prove it's correct — and living with the consequences in the field. If you've ever been frustrated by a "good enough" camera that quietly poisoned your perception pipeline, this is the role where you get to build the thing correctly.

What you'll own

• Hands-on embedded and board-level experience — required. You've built on Raspberry Pi and NVIDIA Jetson Orin-class compute, and you're comfortable at the PCB level: reading schematics, bring-up, interfacing sensors, and debugging hardware rather than treating the board as a black box.
• Design and build a custom stereo camera end to end — sensor selection, optics, mechanical alignment, and the trade-offs between coverage, distortion, and geometric fidelity.
• Architect synchronized dual-camera capture with hardware-level, frame-accurate triggering, so both streams stay aligned in time under real motion — no rolling-shutter smear, no drift between the eyes.
• Fuse vision with high-rate inertial sensing, holding the camera and IMU streams in tight temporal lock so the data is clean enough for robust visual-inertial odometry.
• Own on-device video encoding — efficient, modern compression that keeps high-resolution, real-time streams within realistic bandwidth and storage budgets without throwing away signal.
• Build a metrology-grade calibration pipeline — intrinsics, extrinsics, and temporal calibration — validated to sub-pixel reprojection accuracy and a precisely rectified stereo pair. You define what "calibrated" means here and enforce it.
• Take the device from bench to field — characterize it, break it, quantify its error, and make it reliable enough to trust at scale.

What we're looking for

• Hands-on experience building or bringing up multi-camera / stereo systems, not just consuming their output.
• Strong grasp of camera geometry and calibration — you can rigorously calibrate a stereo + IMU rig and defend the numbers.
• Experience with visual-inertial systems — sensor fusion, VIO, or SLAM in practice.
• Working knowledge of stereo depth / disparity estimation, including modern learned approaches, and dense depth-map generation.
• Comfort with real-time capture pipelines and on-device video encoding — timing, buffering, compression, throughput.
• Fluency in C++ and/or Python, and the pragmatism to profile and optimize where it actually matters.
• A first-principles, ownership mindset — you're comfortable being the person who decides, builds, and is accountable for the result.
• Robotics background — a huge plus. Perception, controls, or systems work on real robots; understanding how capture quality propagates into downstream robot learning.
• Embedded / edge deployment experience (e.g. Jetson-class or similar compute).
• Firmware or FPGA work around sensor timing and triggering.
• Background in robotics perception, AR/VR, autonomous systems, or egocentric capture.
• Experience taking a hardware device from prototype to reliable production use.