Career listing · authored curriculum in review
Data Scientist
Save Data Scientist as an interest and organize future course, credential, and career research around it.
Current availability
The subject-specific curriculum is authored but not yet published for completion. Learners may inspect its scope and learning design now. Start, grading, and completion remain unavailable until every listed gate passes.
- qualified review approval is required
- reviewer identity, qualification, evidence, and approval date are required
What this course establishes
- Frame a decision and estimand from the data-generating process, define units/labels/features and produce a reproducible, privacy-aware data and experiment plan
- Build leakage-safe baseline and candidate pipelines, choose valid splits and metrics, quantify uncertainty, inspect calibration and subgroup errors, and reproduce results
- Specify human oversight, model documentation, deployment tests, drift/performance monitoring, incident response, rollback and accessible communication without overstating causality or fitness
Entry knowledge
- Python or another analytical language
- SQL and tabular data manipulation
- Algebra, functions, probability, descriptive statistics, sampling and confidence-interval basics
Scope boundaries
- Uses fictional data only and does not authorize real automated decisions, sensitive-trait inference, surveillance, re-identification, production deployment, clinical/legal/financial conclusions, or collection without authority and risk review
- Does not replace domain experts, statisticians/causal-inference specialists, privacy/security/legal teams, accessibility research, affected-community participation, independent validation, software/model risk management, or accountable decision owners
Fastest competency route
Completion is self-paced and competency-based. Prior skill can shorten study only when all mathematics, code, validation, governance and revision evidence is demonstrated; no fixed duration, job, certification, model performance, deployment approval, regulatory compliance or business outcome is promised.
- Pass diagnostics for problem framing, sampling, leakage, splits, baselines, metrics, calibration, subgroup analysis, uncertainty, reproducibility, governance and monitoring
- Complete the supplied synthetic service-follow-up prediction analysis and model-risk dossier
- Correct every identity, timing, duplicate, missingness, label, leakage, split, metric, threshold, subgroup, privacy and deployment defect before final assessment
Full self-paced route
- Problem, data-generating process, quality and reproducibility
- Leakage-safe modeling, evaluation, uncertainty and interpretation
- Risk governance, deployment, monitoring and communication
- Applied portfolio and final assessment
Lesson and assessment map
Decision framing, data-generating process, quality, and reproducibility
Define a legitimate scientific question and auditable dataset before choosing an algorithm.
4 questions: recall, application, judgment, and error correction.
Baselines, leakage-safe pipelines, evaluation, uncertainty, and interpretation
Compare reproducible models under a deployment-matched design and select an operating policy based on error costs and uncertainty.
4 questions: recall, application, judgment, and error correction.
Risk governance, deployment, monitoring, incidents, and accessible decisions
Convert a validated model into a controlled sociotechnical system—or a documented no-deploy decision—with accountable monitoring and recourse.
4 questions: recall, application, judgment, and error correction.
Required applied assignment
Reproducible prediction and model-risk portfolio
Using only the supplied synthetic records, produce a reproducible analysis from problem framing through a deployment or no-deploy decision. Preserve the source dataset, log every transformation and never represent a model as approved for consequential real-world use without qualified domain, legal/privacy/security, accessibility, statistical, affected-stakeholder and accountable-owner review.
- Decision/impact framing, data-generating-process diagram, prediction-time/label/feature contracts, data sheet, consent/retention map, quality report and leakage register
- Versioned code/environment, grouped temporal splits, simple baseline and candidate pipelines, fold/test evidence, uncertainty, calibration, thresholds, subgroup/time/error and sensitivity analyses
- Model/data card, human oversight and recourse, privacy/security/fairness/accessibility risk register, independent review plan and prohibited-use boundaries
- Acceptance tests, shadow/canary/rollback/kill-switch and monitoring/incident plans, accessible decision memo, peer red-team response and substantive revision history
Passing standard: 85 of 100 points, with substantive revision required.
Continue after this course
- Linear algebra, calculus, probability, statistical inference, Bayesian methods, optimization and causal inference
- Feature stores, distributed processing, orchestration, model registries, containers, observability and ML systems
- Natural-language, vision, time-series, recommender, anomaly-detection and operations-research specializations
- Privacy engineering, security, fairness/impact assessment, human factors, accessibility and sector governance