Website:
ulook.space
Job details:
About the Company
ULOOK is a deep-tech company pioneering an Autonomous RF Signal Intelligence Stack from orbit. We develop autonomous satellite swarms and advance RF sensing technologies which transform faint electromagnetic emissions into real-time intelligence. Utilising our proprietary technology we identify, geolocate and fingerprint complex signals in environments where traditional sensors fail. We operate at the intersection of defense, space, and intelligence, providing global partners with advanced RF, maritime, and aerial surveillance capabilities, strengthening security and situational awareness across all domains.
About the Role
We are looking for a Machine Learning Engineer who has gone beyond training and fine-tuning models and has hands-on experience in designing and developing ML architectures from the ground up. This role demands a deep understanding of the mathematical frameworks that underpin modern machine learning, from gradient-based optimisation and loss surface geometry to probabilistic inference and regularisation theory. You will not simply consume APIs or call library functions. You will derive, implement, and validate models where every architectural choice is rooted in first principles. Equally important is the ability to take those models to the edge. You will compress, quantise, and optimise ML workloads for deployment on resource-constrained hardware, including satellite and airborne platforms, ensuring that inference quality is preserved under strict compute, memory, and power budgets. This role sits at the core of ULOOK’s intelligence pipeline, bridging deep mathematical modelling with real-world, on-device deployment.
Responsibilities
- Design and develop custom ML model architectures tailored to RF signal classification, emitter fingerprinting, and anomaly detection.
- Formulate problem-specific loss functions, optimisation strategies, and training regimens informed by the statistical and geometric properties of the data.
- Analyse model behaviour through the lens of gradient dynamics, convergence theory, generalisation bounds, and information-theoretic metrics to diagnose issues and guide architectural refinements.
- Own the full model lifecycle from mathematical formulation and prototyping through training, validation, and deployment-ready packaging.
- Develop and execute Edge AI deployment strategies, including post-training quantisation, quantisation-aware training, weight pruning, knowledge distillation, and mixed-precision inference to meet the constraints of spaceborne and embedded platforms.
- Benchmark and profile models across target hardware, characterising trade-offs between bit-width reduction, latency, throughput, memory footprint, and task accuracy.
- Collaborate with embedded and RF systems engineers to integrate optimised models into real-time inference pipelines with deterministic latency and minimal resource usage.
- Work closely with signal processing and numerical algorithms teams to align feature extraction, data representation, and model input pipelines.
- Maintain rigorous experiment tracking, model versioning, and reproducibility practices across all research and deployment workflows.
Qualifications
- Bachelor’s degree in Computer Science, Electrical Engineering, Mathematics, Physics, or a related quantitative field.
- 3 to 5 years of hands-on experience in machine learning, with demonstrable work in developing model architectures, not limited to training or inference using existing frameworks.
- Strong mathematical foundation in linear algebra, calculus, probability theory, optimisation, and statistical learning theory as applied to ML.
- Practical experience with at least one Edge AI deployment workflow, including quantisation (INT8, INT4, mixed-precision), pruning, or knowledge distillation using tools such as TensorRT, ONNX Runtime, TFLite, or equivalent.
- Proficiency in Python with deep experience in PyTorch or TensorFlow at the lower abstraction layers (custom layers, autograd functions, training loops).
- Working knowledge of CUDA and GPU-accelerated computing, including experience writing or optimising custom CUDA kernels, understanding GPU memory hierarchies (shared, global, registers), thread and block execution models, and profiling GPU workloads.
- Demonstrated experience taking at least one model from original formulation through implementation, training, and validation on real-world or experimentally controlled data.
- Comfortable reasoning about numerical stability, floating-point behavior, and precision artifacts in training and inference, including FP16/BF16 mixed-precision dynamics and CUDA-level numerical considerations.
Preferred Skills
- Master’s degree or advanced coursework in machine learning, deep learning, signal processing, or applied mathematics.
- Experience deploying ML models on embedded GPUs, FPGAs, NPUs, or microcontroller-class hardware.
- Experience with CUDA C/C++ beyond wrapper-level usage, including kernel-level memory optimisation, stream-based concurrency, and integration with cuDNN, cuBLAS, or CUTLASS libraries.
- Familiarity with model compilation and hardware-aware optimisation tools such as Apache TVM, NVIDIA TensorRT, Qualcomm AIMET, or OpenVINO.
- Exposure to RF signal processing, communications, or spectral analysis domains.
- Experience with on-device learning, federated approaches, or adaptive inference techniques for non-stationary environments.
- Familiarity with MLOps practices, including CI/CD for model pipelines, experiment tracking (MLflow, W&B), and containerised deployment.
- Proficiency with Git and collaborative software workflows.
Why Join ULOOK?
- Own the machine learning engine at the heart of a next-generation RF intelligence platform.
- Work on hard technical problems that directly impact India’s strategic space and defence capabilities.
- Collaborate with a multidisciplinary team from leading space, defence, and research institutions.
- Build models that will run on real satellites and airborne platforms, not just cloud GPUs.
Click on Apply to know more.