Senior Data & AI Scientist

This is a fully remote position, open to applicants in Georgia, +1 more country.

📋 Description

• Train, fine-tune, and assess transformer-based sequence models to predict behavior using extensive trading data.

• Create feature representations and embedding strategies for structured, event-sequence behavioral signals.

• Model comprehensive multi-horizon behavioral distributions, which include activity, volume, PnL, exposure, and hold time, instead of just single-point or label predictions.

• Establish and surpass a stringent baseline ladder.

• Develop leakage-safe, chronologically split evaluations that validate claims of improvement.

• Enhance large-scale GPU training pipelines on Vertex AI, focusing on throughput, memory, data loading, checkpoint/resume capabilities, and cost efficiency.

• Oversee the entire modeling lifecycle, encompassing data handling, representation learning, training, large-scale experimentation, and thorough evaluation.

• Collaborate with psychologists, bioinformaticians, AI engineers, and researchers to transform theoretical concepts into production-ready AI systems.

⛳️ Requirements

• Over 6 years of experience in Data Science, Machine Learning, or AI.

• Practical experience in training deep sequence models (Transformers/attention) and/or fine-tuning foundational models.

• Strong proficiency in PyTorch and contemporary deep learning workflows; knowledge of Lightning is a plus.

• Experience in distributional and representation learning, focusing on modeling latent structures and full distributions rather than solely labels.

• Experience in designing, executing, and analyzing large-scale ML experiments.

• Proven ownership of the entire modeling lifecycle, from data preparation to training, evaluation, and iteration.

• Expert-level proficiency in Python.

• Deep understanding of deep learning, representation learning, and statistical modeling.

• Familiarity with high-dimensional and event-sequence/temporal datasets.

• Ability to create robust evaluation methodologies, including chronological/held-out splits, leakage control, gap purging, and honest baseline comparisons.

• Comfortable with cloud GPU training and real-world failure scenarios, including throughput bottlenecks, OOM issues, dataloader/prefetch challenges, checkpoint/resume processes, and cost management.

• Experience with zero-inflated, heavy-tailed, or hurdle modeling and calibration for sparse targets is beneficial.

• Knowledge of world-model or model-based/latent-transition (JEPA-style) modeling is advantageous.

• Familiarity with financial, trading, or high-frequency behavioral data is a plus.

• Background in behavioral science, psychology, economics, or cognitive modeling is beneficial.

• Experience with distributed GPU training frameworks, including FSDP, DeepSpeed, Accelerate, or DDP, is a plus.

• Familiarity with LoRA, QLoRA, PEFT, RLHF, or other techniques for adapting foundation models is an advantage.

• Publications or contributions to open-source projects related to sequence models or foundation models are advantageous.

🏝️ Benefits

• Benefits provided by SD Solutions staffing company; specific benefits are not outlined in the posting.