The problem we're solving

GPS fails indoors. The alternative — bluetooth beacons, WiFi fingerprinting, UWB anchors — is expensive, unreliable, and doesn't scale. We took a different approach: we use the Earth’s magnetic field.

We utilize the unique magnetic fields at each location in a building, affected by steels and concrete surrounding it. Combined with inertial sensor data and learned motion models, this is enough to position a person inside a mall, hospital, or transit hub with meter-level accuracy — no installed hardware required. That's Oriient.

We've proven the concept. Now we're pushing the performance envelope: tighter accuracy, faster convergence, more robust behavior across diverse device hardware and building types. That's where you come in.

What you'll work on

You'll be one of a small number of engineers who own the core positioning engine — from research to production:

• Designing and improving localization algorithms: particle filters, Kalman variants ,learned motion models
• Sensor fusion across IMU, magnetometer, and barometer signals, accounting for device heterogeneity and real-world noise
• Translating ideas from academic literature into working prototypes, then into production code running at scale
• Building evaluation infrastructure to benchmark algorithm changes against large real-world datasets
• Investigating failure modes — hunting down the "why" behind edge cases and anomalies in the field

The work is genuinely end-to-end. You'll go from reading a paper to watching your change improve accuracy across thousands of real buildings.

What we're looking for

• B.Sc. in Electrical Engineering, Applied Math, or a related field
• 3+ years working on estimation, signal processing, or sensor fusion problems — particle filters, Kalman filters, or similar
• Strong mathematical footing: you can read an academic paper, identify what's useful, and build a working implementation
• Proficient Python; MATLAB familiarity is useful for rapid prototyping
• You care about correctness and you're patient with ambiguity — real-world signals are messy
• Nice to have: experience with deep learning for time-series or positioning, exposure to cloud infrastructure or large-scale data pipelines.

What's true about working here

• Small team, real ownership. You'll see the direct impact of your work on product behavior.
• The technical stack is unusual: geomagnetic positioning, IMU fusion, and learned models aren't typical ML infrastructure problems.
• We're deployed in large retail chains, hospitals, and campuses — scale isn't hypothetical.
• Flat structure, no lengthy approval chains. If you have a better approach, you can build it and test it.
• We're in a growth phase, which means the problems are expanding faster than the team — there's room to carve out areas of deep ownership.