Chief Scientist (NCS/Job/ 3982)

PhD in Applied Mathematics or Statistics — or a closely related quantitative field (Econometrics,
Operations Research, Machine Learning / CS-theory, or a physical science with heavy statistical
methodology). In place of the degree, we will consider an exceptional, demonstrable research record
— peer-reviewed publications, patents, or causal/statistical systems you took to real-world impact.
Deep, first-principles command of causal inference and applied statistics — structural causal
models, DAGs, confounding and conditioning, partial vs. plain correlation, control charts, multivariate
SPC (PCA-based T²/SPE), time-series (stationarity, lag, autocorrelation), and the failure modes of each.
You challenge a result with a calculation, not an opinion, and you know when a method is being overstretched
BONUS
Our DNA — how we work
A track record of taking a method from idea to validated impact — paper or theory, to prototype,
to a result you proved correct against ground truth on messy real-world data. You’ve designed the
validation, not just the model.
The judgment to set scientific direction — to choose the right method for the problem (and reject
the fashionable-but-wrong one), to define what “correct” means for a noisy industrial signal, and to
own that call.
Enough hands-on coding to be self-sufficient — Python and the scientific stack (NumPy, pandas,
scipy, scikit-learn, statsmodels), comfortable pulling data from SQL and running an analysis end to
end. You build rigorous prototypes and can read production code; you do not need to be a
production/MLOps engineer — that’s what the engineering team is for.
Exceptional communication. You can defend a causal argument to a CTO and explain the same
finding to a plant manager in the same week, and write the one-pager that ends with a
recommendation, not a shrug.
Founder’s temperament. You want to build the research function — comfortable with ambiguity, high
autonomy, and a fast-moving startup where the science, the validation and the customer’s problem
often land in the same week.
Causal-discovery depth — constraint/score-based structure learning, Granger / PCMCI, do-calculus,
counterfactual reasoning — at the level where you’ve recovered structure from real time-series and
validated it.
Industrial / manufacturing / sensor data, process control, or autonomous control systems.
Process-engineering or physics intuition — thermodynamics, mass/energy balances, first-principles
plant models.
Optimization (numerical, stochastic, or process-oriented), translating causal insight into actionable
setpoints.
Exposure to LLM-augmented analytics pipelines (RAG, eval frameworks) and a clear-eyed view of their
failure modes.