Senior Data Scientist, Applied ML

Vitol•Houston, TX

1d•Onsite

About The Position

Vitol’s Applied ML team in Houston works directly with trading desks to find commercial value through data and applied data science. Projects begin with questions from traders or analysts, and the role works alongside them from problem framing through production and ongoing support. The position owns both the science and the path to production, with success measured by whether desk users adopt the tools to make better decisions. The role builds LLM, NLP, and multimodal pipelines to extract structured signals from unstructured sources, and develops models for outage impacts, disruption patterns, anomalies, and regime changes in fundamental data. Proven models are packaged, tested, deployed through CI/CD, and supported by monitoring and drift detection. Insights are delivered through tools traders already use, including Teams, alerts, and Vision Excel and Python tooling. The role also evaluates commercial AI tools and communicates results, uncertainty, and limitations to desk users and senior stakeholders. The position is onsite in Houston five days per week. The Applied ML team sits close to trading desks, enabling fast feedback and direct collaboration. The role offers room to choose an approach while expecting ownership of reliable, commercially useful outcomes.

Requirements

At least 5 years applying machine learning and NLP to messy, real-world data, including taking models to production and supporting them after launch.
Experience working directly with commercial users in trading, finance, or a similar fast-paced setting, with examples of decisions influenced by the work.
Master’s or PhD in statistics, mathematics, physics, engineering, or equivalent depth gained in industry.
Strong statistics and experimental design skills.
Hands-on experience building LLM extraction and classification pipelines, including evaluation, prompt and retrieval design, and cost and latency trade-offs.
Python experience with scikit-learn, PyTorch, and HuggingFace, and the ability to write production-quality code that others can review and maintain.
Production engineering experience with Docker, AWS, CI/CD, and MLOps, including model versioning, monitoring, and retraining.
Experience with anomaly detection and combining signals across multiple time-series and text sources.
Ability to explain what a model means for a trade in concise, clear terms and to engage with users who have deeper market knowledge.

Nice To Haves

Knowledge of energy or commodities markets.
Graph modeling experience.
Sentiment analysis experience.
Experience with agentic workflows and MCP.

Responsibilities

Work with traders and analysts to scope projects around market questions, clear commercial hypotheses, and defined go/no-go points.
Review progress with desk users, share intermediate results, test findings against market knowledge, and adjust as signals or priorities change.
Build LLM, NLP, and multimodal pipelines that extract structured signals from unstructured sources such as outage reports, news, filings, and web content.
Design and run rapid experiments across modeling approaches, and stop weak ideas early.
Build models to predict outage impacts, classify disruption patterns, and detect anomalies and regime changes in fundamental data.
Take successful models into production using tested, packaged deployments through CI/CD, with monitoring and drift detection.
Deliver insights through Teams, alerts, and Vision Excel and Python tooling rather than defaulting to standalone dashboards.
Source data through the Vision platform; coordinate with the Vision team for new sources and follow its existing web-scraping architecture.
Evaluate commercial AI platforms and tools against desk problems and provide clear recommendations.
Explain results, uncertainty, and limitations to traders, analysts, and senior stakeholders.
Support production models after launch and make them monitored, documented, and supportable by the wider team.
Document failed experiments and build reusable components, such as extraction pipelines and evaluation harnesses.