Machine Learning Engineer

Weekly Rate: $9,000/week

Overview

The Machine Learning Engineer focuses on model training, optimization, and deployment strategies. This role ensures models are efficiently trained with high-quality datasets, properly validated, and optimized for edge deployment while maintaining accuracy targets.

This specialized position bridges the gap between research and production deployment, ensuring that cutting-edge computer vision algorithms can operate reliably in real-world competition environments while meeting stringent performance and accuracy requirements.

Key Responsibilities

Model Training and Infrastructure - Build automated training pipelines that efficiently process large-scale datasets. Set up scalable training infrastructure using GPU clusters and optimized data pipelines that enable rapid experimentation and model iteration.

Dataset Management and Quality - Curate and manage high-quality training datasets with proper annotation workflows. Implement synthetic data generation techniques to augment training data for edge cases and ensure comprehensive coverage of squat variations.

Model Optimization Techniques - Apply advanced optimization techniques including quantization, pruning, and knowledge distillation to reduce model size and inference latency. Ensure models meet the constraints of edge deployment while maintaining >95% accuracy.

Transfer Learning and Adaptation - Leverage transfer learning to adapt state-of-the-art models for squat-specific detection. Fine-tune pre-trained models on domain-specific data to accelerate development and improve performance.

Edge Deployment Optimization - Optimize models for specific edge hardware using frameworks like TensorRT, CoreML, and OpenVINO. Ensure models run efficiently on NVIDIA Jetson and similar platforms with sub-50ms inference time.

Continuous Learning Systems - Implement active learning pipelines that continuously improve models using production data. Design automated retraining workflows that keep models current with evolving athlete techniques and venue conditions.

A/B Testing and Validation - Develop comprehensive A/B testing frameworks for systematic model version comparison. Validate improvements through rigorous testing before production deployment.

Neural Architecture Search - Explore and implement neural architecture search techniques to discover optimal model architectures for squat detection. Balance accuracy, latency, and model size constraints.

Performance Monitoring and Analysis - Monitor model performance in production environments, tracking accuracy, latency, and resource utilization. Implement alerting systems for performance degradation and anomaly detection.

Bias Detection and Fairness - Implement strategies to detect and mitigate bias in model predictions. Ensure models perform consistently across diverse athlete populations and body types.

Required Skills

Machine Learning Engineering Excellence demonstrates strong experience building production ML systems that handle real-world complexity and scale effectively. They have managed the complete ML lifecycle from data preparation through deployment monitoring, solving practical problems that arise in production environments. Their experience includes diagnosing model performance issues and implementing solutions that maintain reliability under varying operational conditions.

Deep Learning Framework Proficiency combines solid TensorFlow and PyTorch knowledge with practical optimization skills for resource-constrained deployment environments. They understand framework trade-offs and can implement advanced optimization techniques like quantization and pruning to improve deployment performance. Their Python expertise includes performance profiling and memory optimization techniques needed for efficient operation on limited hardware resources.

Edge Deployment Understanding involves comprehensive knowledge of memory constraints, power limitations, and latency requirements that define successful edge device deployments. They have experience optimizing models for hardware platforms where resource efficiency is critical, making informed trade-offs between model accuracy and performance requirements. Their familiarity with TensorRT, CoreML, and OpenVINO helps bridge research advances with practical deployment constraints.

MLOps and Production Practices includes building automated pipelines that handle model training, validation, and deployment with minimal manual intervention requirements. They understand the challenges of monitoring ML model performance in production environments where data distribution changes can impact accuracy over time. Experience with computer vision applications helps them navigate the specific challenges of video processing and pose estimation that have unique requirements compared to other ML domains.

Phase Allocation

The Machine Learning Engineer begins with partial involvement during Alpha phase for initial model architecture design and dataset preparation. Full-time engagement through Beta and Gamma phases focuses on intensive model training, optimization, and edge deployment preparation. The role continues with decreasing allocation through Delta and Full Release phases to support continuous learning and model refinement.

Phase	Weekly Rate	Allocation	Duration
Alpha	$4,500/week	50%	10 weeks
Beta	$9,000/week	100%	12 weeks
Gamma	$9,000/week	100%	8 weeks
Delta	$6,750/week	75%	10 weeks
Full Release	$3,600/week	40%	12 weeks

Deliverables

Training Pipeline Infrastructure. Automated ML training pipelines that handle data ingestion, preprocessing, model training, and validation with full reproducibility. These pipelines support distributed training across multiple GPUs while maintaining experiment tracking and versioning for all model iterations throughout the development lifecycle.

Optimized Models for Edge Deployment. Production-ready models specifically optimized for edge hardware constraints including quantization, pruning, and architecture modifications. These models achieve the required balance between accuracy and inference speed while fitting within memory and computational limitations of deployed edge devices.

Dataset Management System. Comprehensive data infrastructure for collecting, labeling, versioning, and augmenting training datasets with quality control mechanisms. This system ensures data consistency across training iterations while supporting active learning workflows that continuously improve model performance based on production feedback.

Model Performance Reports. Detailed analysis of model accuracy, latency, and robustness across diverse testing scenarios including edge cases and adversarial conditions. These reports provide statistical validation of model performance while identifying failure modes and recommending targeted improvements for specific use cases.

A/B Testing Framework. Production experimentation infrastructure enabling safe deployment and comparison of multiple model versions in live environments. This framework supports gradual rollouts, performance monitoring, and automatic rollback capabilities while collecting metrics for data-driven model selection decisions.

Continuous Learning Pipeline. Automated systems for model retraining based on production data, drift detection, and performance degradation monitoring. These pipelines ensure models remain accurate as competition conditions evolve while maintaining version control and rollback capabilities for all deployed models.

Model Documentation. Comprehensive technical documentation covering model architectures, training procedures, hyperparameter selections, and performance characteristics. This documentation enables knowledge transfer, reproducibility, and informed decision-making about model deployment and optimization strategies.

Deployment Strategies. Detailed plans for model deployment across edge infrastructure including versioning strategies, update mechanisms, and fallback procedures. These strategies ensure smooth model updates without service disruption while maintaining performance consistency across heterogeneous hardware deployments.

Success Criteria

Model Accuracy Achievement. Machine learning models consistently achieve greater than 95% accuracy in squat detection and validation across diverse athlete populations and movement variations. This accuracy is maintained in production environments with real-world lighting conditions, camera angles, and occlusion scenarios encountered during competitions.

Inference Performance Standards. Models achieve sub-50ms inference time on target edge devices while maintaining accuracy requirements, enabling real-time pose estimation within system latency budgets. This performance is validated across different hardware configurations and maintained throughout extended operation periods without degradation.

Model Size Optimization. Deployed models remain under 100MB in size through effective compression techniques while preserving accuracy and performance characteristics. This optimization enables efficient deployment on edge devices with limited storage and memory resources while supporting over-the-air updates.

Edge Deployment Success. Models successfully deploy and operate on target edge hardware including NVIDIA Jetson platforms with stable performance across temperature variations and extended operation periods. Deployment validation includes stress testing under competition conditions with multiple concurrent video streams.

Continuous Improvement Validation. Demonstrated model performance improvement through iterative training cycles based on production data and active learning strategies. Metrics show consistent accuracy gains and reduction in failure modes as models adapt to new scenarios encountered in live deployments.

Edge Case Robustness. Models maintain reliable performance when handling challenging scenarios including partial occlusions, unusual body positions, and equipment interference. Comprehensive testing validates model behavior across identified edge cases with graceful degradation rather than catastrophic failure when encountering novel situations.