Senior Data Scientist

We are Skyward. That is, a love for people, for improvement, for human advancement through information technology. We are a people-centered business with a desire to serve others. We are diverse and unified; creative and collaborative; a collection of complementary, not competing talents. And though on the surface we remain relaxed, beneath, a torrent of energy links us to our civic tech mission.

(CONTINGENT HIRE BASED ON CONTRACT AWARD)

The kind who looks at a tangle of federal and private datasets that don’t share schemas, don’t share IDs, and were never meant to talk to each other, and gets a little excited. The kind who knows that the answer is almost never “throw a bigger model at it” and is almost always “understand the data first, then pick the model.”

The kind who can sit across from a federal subject matter expert and explain what a Leiden community is without making them feel dumb, and without dumbing it down either.

If you’ve ever quietly fixed someone else’s “production” notebook on a Friday afternoon – the one with hard-coded paths, no random seed, and a function called _v3() – this might be you.

Come join us if you're motivated to learn from others, to learn from mistakes, to be part of a future-looking and growth-oriented team.

• Lead end-to-end data science experiments. From a data readiness assessment, through clustering and topological risk modeling, into unstructured-data enrichment and entity resolution
• Run exploratory data analyses (EDAs) on government-furnished data inside a government-controlled environment: profile completeness, find the schema mismatches, flag the gaps, and document what the data can and cannot support before a single model gets trained
• Apply graph analytics. Leiden community detection, betweenness and eigenvector centrality, motif analysis, temporal cluster detection, link prediction. And be able to explain in plain English what each one means and what it doesn’t
• Train interpretable classifiers (logistic regression, gradient boosted trees) and Graph Neural Networks (Graph
  
  SAGE, GAT) where the data supports them; reach for unsupervised anomaly detection when labels are thin (and they will be)
• Run probabilistic entity resolution across biographic, behavioral, and biometric features using tools such as Senzing. Handle name transliteration, DOB variation, and fuzzy address matching like the working scientist you are
• Apply LLM-based Named Entity Recognition and relationship extraction to unstructured field text and quantify whether the extracted edges actually change the graph (rather than just adding noise that looks impressive in a deck)
• Wrangle messy data and recommend supplemental, de-identified data sources that would enrich the analysis, and document the case for each recommendation so the customer’s privacy and legal teams can make an informed call
• Document everything so the customer can rerun it after you’re gone. Reproducibility is a hard acceptance criterion here, not a “nice to have.”
• Hold the line on rigor: confidence intervals on everything, null findings documented with the same care as positive ones, and zero appetite for dashboard theater

• An active Secret clearance at time of hire
• 7+ years of applied data science experience, with at least 2 years that included graph analytics or network analysis as a primary tool, not a side dish
• Strong production-grade Python: pandas, Num Py, scikit-learn, networkx (or graph-tool / igraph), and at least one GNN library such as PyTorch Geometric or DGL
• Real-world experience with probabilistic record linkage / entity resolution - Senzing, dedupe, FEBRL, Magellan, or a homegrown Fellegi-Sunter implementation you’d be willing to defend in a code review
• Comfort working without labels. Anomaly detection, positive-unlabeled learning, isolation forests, autoencoders. You know how to evaluate a model when there’s no clean ground truth to compare it to
• Interpretability discipline: SHAP, feature importance, partial dependence plots, and the wisdom to pick the simpler model when it’s the right one
• LLM application experience beyond prompt-and-pray. Entity and…