Senior DataScientist.

Indio builds its own AI and large-scale data products for the energy industry. We don't do client work: our engineers build Indio's products, working directly with the people who use them and alongside our data science team, and what we build goes to production.

We keep the team small and the bar very high. If you're the engineer other engineers go to, we'd like to hear from you. Parts of our work are confidential; we share more in interviews.

You'll take on hard, open-ended modelling problems with large, messy, real-world data, and turn that data into models people rely on for decisions with real money behind them. This is a senior, hands-on role with real ownership: you'll take problems from framing to production, decide what a good answer looks like, prove the model gets there and keep raising the bar.

We're looking for someone exceptional: the data scientist other data scientists go to when a model has to be right, and who can prove it.

• —Automated title. AI that reads deeds, leases, probate and court records and builds chains of title for whole counties: the work a landman does by hand, tract by tract. A confident wrong answer is expensive, so every result has to be measured and explainable.
• —Property tax consulting. Models over appraisal rolls, tax rolls and oil and gas production data that show where properties are valued too high, so we can lower our clients' property taxes.
• —Oil and gas mineral acquisition. Forecasting and valuation models that decide which mineral and royalty interests we buy, with our own money behind every decision.

• —Forecasting under uncertainty. Models on sparse, noisy, irregular time series that hold up out of sample, come with calibrated uncertainty and stay stable as new data arrives.
• —Making unstructured data usable. Large volumes of documents and semi-structured records turned into reliable, structured data, using modern machine learning and language models where they help and simpler methods where they're enough.
• —Agentic AI you can trust. Multi-step language-model workflows that read, reason, check and hand off, held to the same standard as any other model: measured, repeatable and explainable.
• —Bringing imperfect sources together. Data from many sources of uneven quality, with that quality measured and improved instead of assumed.
• —Decisions, not just predictions. Risk and value quantified, so the business can act on the full range of outcomes, not a single point estimate.
• —Creating ground truth. Where no benchmark exists, the labelled data and evaluation that every model is judged against, built with domain experts.

• —Own modelling problems end to end: framing, data, method, evaluation and production
• —Agree measurable success criteria with the business and report honestly against them, including when a model isn't good enough yet
• —Design datasets, labelling guidelines and evaluation sets, and keep them current as the data changes
• —Build evaluation into development, so every model, prompt or data change shows its effect before it ships
• —Design, evaluate and harden language-model pipelines: model choice, structured outputs, guardrails, cost and latency
• —Choose the right method for each problem, from statistics and classical machine learning to deep learning and language models, and back the choice with evidence
• —Write production-quality Python alongside our engineers, and explain results and uncertainty clearly to leadership, analysts and engineers
• —Help set the standards and the hiring bar for data science as the team grows

• —Evaluation first: every model, prompt and data change measured before it ships, and monitored after
• —The simplest method that works, backed by evidence
• —Ground truth built with domain experts where no benchmark exists
• —AI coding assistants used every day, with their work checked
• —Tests, reviews and documentation as part of the work, not an afterthought

We hire on evidence, not titles or keyword lists. Every claim on your resume will be discussed in depth in a live, camera-on interview, so list only work you did yourself and can explain in detail.

• —6+ years in data science or applied machine learning, including several models you shipped that people relied on for real decisions
• —Measurable impact, with the numbers: an error rate you cut, a forecast you made more accurate out of sample, or manual work you automated
• —An evaluation you designed from scratch: the metric, the labelled data, the baseline and the decision it changed
• —Forecasting or time-series modelling on messy real-world data, including backtesting and quantified uncertainty
• —Language-model or agentic systems in production, with quality you measured and improved, not just demoed
• —Algorithm design beyond library defaults: you've built or adapted a method when the standard one wasn't good enough, and proved the new one was better
• —Breadth across tabular, time-series and text data, and the judgement to know which methods fit which
• —Strong statistics: you know when a result is noise, and you can explain why
• —Strong Python and SQL, clean tested code and comfort working in a production codebase
• —Clear written communication in English

• —Geospatial data and spatial analysis
• —Bayesian methods, probabilistic forecasting or calibration
• —Graph databases or graph-based modelling
• —Regulated or high-stakes settings where every result must be explainable
• —An MS or PhD in a quantitative field, published research, or top results in competitions such as Kaggle, math olympiads or ICPC, or equivalent depth shown through your work

• —Real data at scale. Proprietary, real-world data with long history, not another clickstream.
• —Hard problems. Open questions with no off-the-shelf answer, where your methods can set the standard.
• —Direct impact. Your models feed decisions with clear financial outcomes, and you'll see the results.
• —Experts beside you. Domain specialists who can tell you what's right and why.
• —Ownership. A senior seat with a direct line to leadership and a real say in the roadmap.

1. —Intro call (30 min, camera on): your background and what you want to work on next
2. —Deep dive (90 min): walk us through one model you shipped (the data, the metric, the baseline, the trade-offs and what broke), then design a solution to one of our real problems with us, without code
3. —Practical exercise (about 2–3 hours): a realistic dataset and an open question. We want to see how you frame it, evaluate it and explain it. AI tools are allowed
4. —Live review (45 min): you walk us through your exercise, and we change the question and work through it together
5. —Meet the team (45 min) with engineering, leadership and the domain experts you'll work with
6. —References from people who saw your work in production

Apply with your resume, a link to work we can inspect and a short written answer about one model you built. The link can be GitHub, a paper, a technical write-up or a Kaggle profile. In your answer, give the problem, the metric, the baseline you beat, how you tested it on data it hadn't seen, and what changed because of it. Applications that list tools without results won't move forward, and we'll ask you to defend every number in the interview. We also require a copy of any thesis or dissertation you wrote at university; we'll ask for it after we review your application, and a redacted version is fine.