Data Scientist

Deviare; South Africa; Vor Ort; ZAR 700.000 - 900.000; Vollzeit

Zusammenfassung

Deviare is seeking an experienced Data Scientist or Psychometrician to validate the platform's proprietary assessment algorithms behind a skills-science platform. You will ensure question banks are statistically reliable before production, requiring a strong statistical background, hands-on NLP validation, and a rigorous QA mindset focused on data accuracy and fairness.

In this gatekeeper role, you stress‑test models, verify data flows from test submissions to rate cards and talent matches, and

Qualifikationen

• Background in data science, data analysis or psychometrics.
• Experience evaluating assessments, surveys or standardised tests.
• Familiarity with Item Response Theory and Cronbach’s alpha.
• Ability to validate NLP models and text-extraction logic.

Aufgaben

• Stress-test core mathematical and logical models to ensure accuracy.
• Validate NLP/text-processing for bias and hallucinations.
• Verify data flow from test submission to rate-card and talent matching.
• Collaborate with C#/.NET backend engineers to report anomalies.

Kenntnisse

Data science

Data analysis

Psychometrics

Statistical testing

NLP validation

QA mindset

Tools

Python

R

SQL

Jobbeschreibung

We are seeking an experienced Data Scientist or Psychometrician to validate the proprietary assessment algorithms behind a skills-assessment and talent-matching platform. The role ensures the platform’s question banks are statistically reliable and of high quality before production launch. It requires a strong statistical background, hands‑on experience validating NLP or text‑extraction logic, and a quality‑assurance mindset focused on data accuracy and fairness.

Context & Impact

The platform assesses candidates and analyses their resumes to produce SFIA‑based skill levels, recommendations, rate cards and talent matches. Its outputs feed directly into hiring and resourcing decisions, so the scoring must be accurate, fair and defensible.

The role acts as the gatekeeper for data quality and algorithmic accuracy. The work involves stress‑testing the core mathematical and logical models, including text processing, resume parsing and scoring. Bias, hallucinated outputs and statistically weak questions need to be found and fixed before launch. The role also verifies the full data flow, from a candidate’s test submission through to the rate‑card and talent‑matching features in the Tenant Portal. It works closely with the C#/.NET backend engineering team to report anomalies and refine the scoring logic.

Why This Role Is Exciting

• Validation before launch: shape the accuracy and fairness of a scoring engine before it reaches production.
• Applied psychometrics: apply Item Response Theory, difficulty and discrimination indices, and reliability measures to identify weak or invalid questions across the question banks.
• NLP edge‑case testing: build edge‑case datasets that test resume parsing and text processing for bias, misclassification and hallucination.
• Testing at scale: design simulation scripts and test datasets that stress‑test the scoring engine at volume.
• Influence on the algorithm: recommend weight adjustments and scoring refinements directly to the engineering team.

What We’re Looking For

The ideal candidate has a background in data science, data analysis or psychometrics. They have strong experience evaluating the quality of assessments, survey data or standardised tests.

Core strengths include:

• Statistical testing for assessment reliability and validity, such as Item Response Theory and Cronbach’s alpha
• Proven experience validating NLP models, text‑extraction tools or similar logic
• Proficiency in Python, R or SQL for analysing database outputs and writing test scripts
• The ability to understand complex business rules and scoring rubrics quickly

The right person has a sharp QA mindset and notices data inconsistencies that standard software testing misses. They work through ambiguity methodically and back their findings with evidence. They are comfortable challenging the logic of a model and explaining statistical findings clearly to engineers. Familiarity with the SFIA framework is useful, but the ability to learn a new domain quickly matters more.