Company Overview

At ALX, we’re unlocking the future we want to see. We’re catalyzing the transformation of Africa, by developing the next generation of bold, innovative, ethical and entrepreneurial leaders. We’re unlocking the potential of the world's largest workforce. The future is calling: be the answer.

Role Summary

Project A is ALX’s AI learning platform. Under it sits a competency model and the hard algorithmic question of the whole system: given what we know about a learner, what should they do next? The Senior AI/ML Engineer designs and builds that engine. It is a probabilistic path-recommendation problem in the family of Bayesian Knowledge Tracing and Knowledge Space Theory, and it has to work from a cold start: no behavioural data yet, so the model is the prior. You encode the prerequisite structure of the domain and let it update as learners come through. We deliberately want an engineer with real modelling depth rather than a pure data scientist on a team this small, the high-value work on day zero is building, not analysing data we don’t yet have.

Specific Responsibilities

Competency Navigation Algorithm

• Own the competency navigation algorithm behind the Learner-Competency-Mapper, its design, implementation, and update dynamics as real data arrives.
• Encode the prerequisite structure of the domain and its priors, so the model performs from a cold start and improves as learners flow through.

LLM Processing & ML Growth

• Build pipelines that turn unstructured platform data into signal - first, a constrained LLM-as-judge answering whether Chidi is effective, with model selection, eval design, and awareness of judges’ own failure modes.
• As the platform accumulates a feedback stream, builds behavioural and at-risk profiling and the models that evaluate learners for the Grader-Competency-Pulser; the role grows into genuine ML/data-science work as the data asset does.

Skill Requirements - Essential

• Probabilistic / Bayesian modelling: real depth — you have designed models from domain structure, not just fit them to data.
• Python & shipping: strong Python and the ability to ship what you design, production pipelines, not notebooks.
• Evaluation: eval design experience, or the judgment to build it fast.

Desirable (not required):

• BKT/KST or psychometrics exposure; MLflow or similar experiment tracking; knowledge graphs. No prior EdTech required but useful.
• Serious probabilistic modelling of structured domains (recommenders, knowledge graphs, causal inference) is great.

Essential Traits for Success

• You reason carefully about your assumptions — in a cold-start model, bad priors compound silently, and you find that problem interesting.
• You learn unfamiliar domains fast and enjoy it.
• You can talk about a model that was wrong and how you found out.

• Own the competency navigation algorithm behind the Learner-Competency-Mapper, its design, implementation, and update dynamics as real data arrives.
• Encode the prerequisite structure of the domain and its priors, so the model performs from a cold start and improves as learners flow through.
• Build pipelines that turn unstructured platform data into signal - first, a constrained LLM-as-judge answering whether Chidi is effective, with model selection, eval design, and awareness of judges’ own failure modes.
• As the platform accumulates a feedback stream, builds behavioural and at-risk profiling and the models that evaluate learners for the Grader-Competency-Pulser; the role grows into genuine ML/data-science work as the data asset does.

• Probabilistic / Bayesian modelling
• Python
• Shipping production pipelines
• Evaluation design
• BKT/KST or psychometrics exposure (desirable)
• MLflow or similar experiment tracking (desirable)
• Knowledge graphs (desirable)
• Serious probabilistic modelling of structured domains (recommenders, knowledge graphs, causal inference)

• BA/BSc/HND
• Real depth in probabilistic / Bayesian modelling
• Strong Python skills
• Ability to ship production pipelines
• Evaluation design experience or judgment to build it fast
• BKT/KST or psychometrics exposure (desirable)
• MLflow or similar experiment tracking (desirable)
• Knowledge graphs (desirable)
• Serious probabilistic modelling of structured domains (recommenders, knowledge graphs, causal inference)