Senior Machine Learning Engineer (Sports Tech / Edge AI) Warsaw

<p dir="ltr"><strong>About Softeq:</strong></p><p dir="ltr">Established in 1997, Softeq was built from the ground up to specialize in new product development and R&amp;D, tackling the most difficult problems in the tech sphere. Now we've expanded to offer early-stage innovation and ideation plus digital transformation business consulting. Our superpower is to deliver all of this under one roof on a global scale.</p><p dir="ltr">We are looking for a hands-on <strong>Senior Machine Learning Engineer </strong>to spearhead the development of an on-device AI solution for sports analytics. You will architect, train, and deploy lightweight, high-performance models that process dual-leg sensor data (IMU) to recognize complex movement patterns in real-time. This is a pure engineering role requiring deep expertise in time-series analysis and edge optimization.</p><p dir="ltr"></p><p><em>We are expecting to grow our team and begin new projects in the next 1-2 months, As such, we’re starting to accept resumes and process chosen candidates.</em></p><p><em>Feel free to apply!</em></p><p dir="ltr"></p><p dir="ltr"><strong>Location:</strong><br/></p> <p dir="ltr"><strong>Warsaw, Poland (B2B contract, fully remote)</strong><br/></p><p></p><h3 dir="ltr"><strong>KEY SKILLS AND REQUIREMENTS</strong></h3><h4 dir="ltr"><strong>1. ML Architectures &amp; Time Series</strong></h4><p></p><p dir="ltr">Deep Learning for Sequences: Deep understanding of modern architectures for time-series processing, specifically:</p><p dir="ltr"></p><p dir="ltr">TCN (Temporal Convolutional Networks): Dilated 1D Convolutions, Residual blocks, Causal padding.</p><p></p><p dir="ltr"></p><p dir="ltr">RNN Variants: Bi-directional LSTM / GRU, layer stacking.</p><p></p><p dir="ltr"></p><p dir="ltr">Hybrid / Attention Models: 1D-CNN + Attention mechanisms (Transformer-lite), Projection heads.</p><p></p><p dir="ltr"></p><p dir="ltr">Classical ML Baselines: Experience with Random Forest and XGBoost based on strong feature engineering (windowed stats, spectral energy).</p><p></p><p dir="ltr"></p><p dir="ltr">Metric Design: Ability to design robust evaluation metrics (Macro-F1, Confusion Matrix analysis) and handle severe Class Imbalance in real-world datasets.</p><p></p><ul> </ul><h4 dir="ltr"><br/></h4><h4 dir="ltr"><strong>2. Model Optimization &amp; Edge Deployment</strong></h4><p></p><ul> <li dir="ltr">Optimization Techniques: Hands-on experience compressing models for mobile:<br/>Quantization: Post-training quantization (PTQ) to INT8.<br/>Pruning: Structured pruning of convolutional and recurrent layers.=<br/>Knowledge Distillation: Training lightweight "student" models based on heavy "teacher" models.</li> <li dir="ltr">Deployment Stack:<br/>Interoperability: Expert-level knowledge of the ONNX ecosystem (export, validation, versioning, opset compatibility).<br/>Mobile Runtimes: Experience preparing models for Core ML (iOS), TFLite / NNAPI (Android), and ONNX Runtime.<br/>Constraint Management: Proven ability to optimize models for strict hardware constraints: Inference &lt; 50–80ms, Model Size &lt; 5–10MB.</li> </ul><h4 dir="ltr"><br/></h4><h4 dir="ltr"><strong>3. Signal Processing &amp; Data Handling</strong></h4><p dir="ltr"></p><p dir="ltr"><br/></p><p dir="ltr">Sensor Data (IMU): extensive experience working with raw accelerometer and gyroscope data (6-axis / 9-axis) and understanding motion physics.</p><p></p><p dir="ltr">DSP Techniques:</p><p dir="ltr"></p><p dir="ltr">Sensor Calibration &amp; Gravity removal.</p><p></p><p dir="ltr"></p><p dir="ltr">Resampling &amp; Synchronization (NTP time sync alignment).</p><p></p><p dir="ltr"></p><p dir="ltr">Normalization techniques (Min-Max, Z-score per session).</p><p></p><p dir="ltr"></p><p dir="ltr">Feature Extraction: RMS energy, Jerk, Spectral Centroid.</p><p></p><p dir="ltr"></p><p dir="ltr">Data Augmentation (Time-Domain): Implementation of Time-warping, Jittering (Gaussian noise), Random window shifts, and Channel dropout.</p><p dir="ltr"></p><p></p><ul> </ul><h4 dir="ltr"><strong>4. Engineering &amp; MLOps</strong></h4><p dir="ltr"></p><p dir="ltr"><br/></p><p dir="ltr">Core Stack: Production-quality Python, expert proficiency in PyTorch or TensorFlow.</p><p></p><p dir="ltr"></p><p dir="ltr">Infrastructure: Experience managing cloud training environments (AWS/GCP), GPU resources, and Docker for reproducible training.</p><p></p><p dir="ltr"></p><p dir="ltr">Validation Strategy: Implementation of strict Subject-exclusive validation schemes (preventing specific user data leakage into test sets).</p><p></p><p dir="ltr"></p><p dir="ltr">Data Pipelines: Building pipelines for multimodal data synchronization (Video + Sensor timestamps) and automated window slicing.</p><p></p><p dir="ltr"></p><p dir="ltr">Tooling: Proficiency with experiment tracking tools (e.g., MLflow, Weights &amp; Biases) to benchmark multiple architecture iterations.</p><p dir="ltr"></p><p></p><ul> </ul><h4 dir="ltr"><strong>5. Soft / L