Senior Machine Learning Engineer (Multimodal AI)

Що робити

Take ownership of the existing multimodal emotion prediction model: master its architecture and limitations, and drive accuracy improvements, particularly on underrepresented emotion classes
Design, train, and evaluate new models on the roadmap: brand fluency/recognition, emotional intensity, saliency, and social ad performance prediction
Bring experience-based judgment to model strategy: assess ideas quickly, select the highest-value experiments, and protect the team from costly dead ends in training time and GPU spend
Build and improve ML infrastructure: migrate training workloads to AWS SageMaker (or Lightning AI), establish proper dataset management, and move from aggregated data snapshots to respondent-level training data via direct database integration
Extend the models with new capabilities: speech understanding encoders, OCR, and LLM-based metadata feature extraction

5+ years of hands-on ML engineering experience, including training and fine-tuning deep learning models end to end (beyond consuming pre-trained APIs or LLMs)
Strong PyTorch expertise
Practical experience with multimodal architectures — video, audio, and fusion/ensemble models (e.g., VideoMAE, ViT, BEATs, HuBERT, CLIP-class encoders)
Solid computer vision background and experience with video data pipelines (frame sampling, feature extraction and pre-caching, large-scale video datasets)
Proven transfer learning and fine-tuning experience: selective layer unfreezing, handling class imbalance and label scarcity