About the role
About the Role
Build, tune, evaluate and serve our own small language and ML models in-house. Role will own the science end to end, base model to production, with the evals and model cards to prove it.
Key Responsibilities:
• Train, fine-tune and distil models—genAI (LoRA/PEFT, RLHF, prompt-based) and classical (transfer learning)—for live product use cases.
• Design evaluation, model tracking and model cards: rubrics, LM-as-judge, regression, drift.
• Own the "tune vs. swap the base model" call; partner with platform to deploy, serve and maintain foundry models.
• Set method and mentor the DS/ML team.
Role Proficiency:
• Fine-tuning/PEFT, transfer learning, RLHF, prompt tuning; evaluation and experiment tracking.
• Strong probability and statistical inference—distributions, Bayesian methods, significance, calibration, experiment/A-B design; and sound handling of skewed/imbalanced data (resampling, class weighting).
• Some MLOps— deploy/serve/monitor (SageMaker, MLflow, Docker; vLLM/Triton a plus).
• MS/PhD in CS/Stats preferred. Fluent in Python/Jupyter, PyTorch, Hugging Face; SQL/Postgres (pgvector), a vector store, a warehouse (Snowflake/BigQuery); Spark or Ray; Airflow/dbt.
• Shipped many models in a large org; strong written and verbal communication skills (tech blogs, publications, talks at meetups/conferences).
Qualifications:
Post-graduate degree (MBA, MTech, LLM, or equivalent) from a top-tier institution.
6+ yrs building ML/AI in production, deep in model training—genAI or classical (at least one; both preferred)—with 3+ yrs leading or tech-leading.
Good to Have:
GenAI product experience (e.g. enterprise RAG bot); text, voice and vision (VLM) models; startup / fast-paced background