Dirac Labs

About Dirac Labs Dirac Labs is building diamond NV-center magnetometers and related quantum sensors that unlock precise navigation when GPS is unavailable. Our work is supported by NASA, NOAA, Emergent Ventures, and the USISTEF. Our mission is simple: provide trustworthy navigation in every location without worrying about spoofing or jamming. Why this role exists We are creating foundation models for magnetic navigation. The goal is a unified model that consumes streams of magnetic data, actively denoises them, matches to prior maps, and outputs one’s location in real time. What will you do Design and train denoising models for magnetic signals across diverse environments and platforms. Build map-matching pipelines that align live magnetic observations with prior magnetic maps to infer pose and position. Architect a unified model that fuses temporal magnetic data, inertial cues, and contextual priors to deliver real-time location with uncertainty bounds. Create data tooling for collection, labeling, augmentation, and synthetic generation of magnetic trajectories. Own the evaluation stack: metrics, offline replay, A-B experiments, on-device latency and power profiling. Productionize models on embedded and edge targets, integrating with our flight computer and navigation stack. Collaborate closely with quantum hardware, firmware, and field ops to close the loop between model assumptions and real-world behavior. Publish internal specs, ablation studies, and decision records that make the system understandable and maintainable. You might be a fit if you have 2+ years in applied ML or a strong research background with shipped systems. Depth in sequence modeling or sensor fusion. Examples include Transformers for time series, state space models, Kalman filtering variants, or diffusion-style denoisers. Experience building real-time inference systems. Comfort with quantization, pruning, and profiling. Strong Python and one systems language. Preferably C++. Practical instincts around data. From designing field runs to cleaning, aligning, and stress-testing datasets. Clear writing and a habit of turning experiments into decisions. Nice to have Prior work on SLAM, map-matching, or PNT. Embedded or edge ML deployment on NVIDIA Jetson, Qualcomm, or ARM targets. Time series foundation models, contrastive pretraining, self-supervised learning at scale. Synthetic data generation, domain randomisation, or physics-informed learning. Experience with inertial sensors, magnetometers, or geophysics datasets. Location Bengaluru, India. In-person by default. Occasional field testing. Compensation Competitive salary, meaningful equity, and the chance to define the company’s navigation intelligence from day one.