AI Scientist

Job description

ABOUT THE SCIENCE TEAM At the heart of Maki People, the Science team is shaping the future of hiring through innovation, rigour, and collaboration. Led by our Head of Science, Aiden Loe, and working closely with our COO, Paul-Louis Caylar, this team drives the development of high-quality content that sets our platform apart. We don’t just create and validate assessments—we innovate. Our work spans: - Expanding a cutting-edge library of tests and tools. - Designing bespoke activities and experiences for clients. - Evaluating and refining AI-driven scoring algorithms and large language models (LLMs) to ensure fairness, accuracy, and transparency. - Leveraging psychometric expertise to build reliable, valid, and impactful assessments. - Developing tools that analyze candidate and job data to predict performance and potential with precision. - Supporting clients in using assessment data to optimize their workforce strategies, from talent acquisition to development and retention. - Leading original studies to explore emerging psychological and technological trends and sharing insights through publications, presentations, and client reports. - Collaborating with regulatory bodies and industry leaders to establish new standards in ethical AI use and hiring practices. - Equipping internal teams and clients with the knowledge and skills needed to understand and apply psychological and AI-driven insights effectively. As Maki continues to grow, the Science team is central to understanding user experiences, refining assessments, and driving broader adoption—all while upholding the highest scientific standards. Your impact as a AI Scientist will go beyond day-to-day responsibilities— you’ll be a key partner in shaping the future of recruitment while driving exceptional outcomes for our clients. ABOUT THE ROLE The AI Scientist works at the intersection of psychometrics, AI, and research, ensuring that Maki’s automated scoring systems are scientifically robust, fair, and continuously improving. 1. SELECTION & EVALUATION OF AI & PSYCHOMETRIC MODELS - Assess the statistical accuracy and reliability of LLMs used for automated scoring (e.g. structured grids, job-specific skills, multilingual proficiency — written and spoken) - Evaluate and calibrate psychometric models (e.g. CTT, IRT, CFA) to ensure scientific validity and comparability across populations and test forms - Test model robustness, including sensitivity to prompt variation, input noise, and adversarial behaviour (e.g. prompt injection) 2. HUMAN-AI COMPARISON & HYBRID EVALUATION MODELS - Design research comparing AI-scored assessments with expert human judgments to ensure validity and alignment. - Benchmark multiple LLMs (both closed- and open-source) across diverse assessment types. - Develop hybrid scoring pipelines combining human oversight and AI-driven analytics. 3. BIAS & FAIRNESS ANALYSIS - Detect and analyse potential biases in AI-generated or psychometric scores across demographic groups. - Apply fairness and bias-mitigation techniques (e.g., reweighting, calibration, subgroup analysis) while maintaining model performance integrity. - Contribute to internal fairness dashboards and compliance documentation, supporting transparent model governance. - Continuously evaluate model generalisability and fairness to ensure all predictive algorithms adhere to ethical and scientific standards. 4. PREDICTIVE ANALYTICS & PERFORMANCE INSIGHTS - Work with large-scale assessment and performance datasets to model relationships between candidate scores, job performance, and retention outcomes. - Develop and test predictive models that estimate success probabilities or identify key behavioural and linguistic predictors of performance. - Collaborate with data science, implementation and customer success teams to translate insights into actionable recommendations for clients and internal stakeholders. 5. ONGOING MODEL MONITORING & ISSUE RESOLUTION - Investigate anomalies raised by clients or internal QA. - Conduct diagnostic analyses and recommend evidence-based improvements. - Implement continuous monitoring systems to track model performance, drift, and stability over time 6. TECHNICAL RESEARCH COMBINING AI & PSYCHOMETRICS - Explore prompt engineering, evaluation methods, and model optimisation techniques - Translate technical findings into actionable insights for non-technical stakeholders. - Contribute to internal and external research (reports, publications, conferences) Eventually as one of the early employee of MakiPeople, you'll be be able to shape the future of the team. We share as much ownership on the way we work and on the product itself as we can as we're convinced our success is 99% due to our team. OUR IDEAL CANDIDATE - Advanced degree (PhD/MSc) in Data Science, Machine Learning, Psychometrics, Computational Linguistics, or Psychology. - Proven expertise in AI model evaluation, psychometric validation, and statistical analysis. - Basic knowledge of psychometric modelling (e.g., IRT, CFA, CAT) and its application in assessment design and validation. - Familiarity with LLMs and NLP techniques used for automated assessment and scoring. - Experience applying fairness and bias testing methodologies in AI-driven decisions. - Skilled in validation research ensuring reliability, construct validity, and practical relevance of assessments. - Proficiency in Python or R and experience with statistical software and cloud databases (e.g., BigQuery). - Strong grounding in ethical AI, data governance, and compliance. - Experienced in collaborating across teams (engineering, product, content) and communicating insights clearly to both scientific and business audiences. - Skilled in data visualisation and research writing, with a track record of publications or applied studies. APPLICATION PROCESS - Stage 1 - Screening assessment (20 mins) - Stage 2 - Hiring manager interview (45 min) - Stage 3 - Power skill assessment with our AI agent (15 min) - Stage 4 - Executive interview (45 min) - Stage 5 - Deep-dive technical interview (60 min) - Stage 6 - Interview with Co-founder (30 min)