The Role
You will build and own the AI and adaptive learning layer of our e-learning products, across teaching, learning, content creation, and analytics, choosing the right technical approach for each problem rather than forcing a single technique.
This is a hands-on role with real ownership. You will be the person who understands this layer best, so you are accountable not just for building it but for making it legible: explaining your decisions, documenting your work, proving its value with evidence, and bringing non-technical colleagues along. The problems are real and substantial: grounding AI in our own educational content, personalizing the learning experience for each student, working across Arabic and English, and doing all of it responsibly with student data and young learners.
What you'll do
- Plan and deliver the AI and adaptive learning features on our roadmap, recommending the best technical approach and sequencing for each.
- Design and build AI features end to end, from prototype to production, drawing on large language models, retrieval, machine-learning models, and classical techniques as each problem demands.
- Own the adaptive learning engine: personalized learning paths, mastery and progress tracking, difficulty adjustment, and targeted remediation, built in tiers that start rule-based and grow more data-driven as real usage data accumulates.
- Define, with content and product teams, the metadata and learning-data foundations that adaptive learning depends on: what each piece of content teaches, at what difficulty, and against which skill or competency.
- Instrument the platform to capture the right learning signals from day one, since data-driven personalization depends on having that data over time.
- Build the systems that ground AI in our own content and data so outputs are accurate, relevant, and safe.
- Stand up and evaluate both self-hosted and API-based models, and decide what to use where based on quality, cost, and privacy.
- Define and enforce responsible-AI practices: safety, guardrails, data privacy, and compliance appropriate to student data and minors.
- Establish evaluation and measurement for every AI and adaptive feature, proving they improve learning outcomes and not just engagement, and report impact and cost in terms leadership can understand.
- Write clean, production-grade Python and integrate AI services with the rest of the platform through well-defined APIs.
- Mentor a junior AI engineer and document the work, so our AI and adaptive learning capability lives in the team and not in one person's head.
Technologies and tools (today's toolkit and the frontier we're building toward)
We are not tied to any single stack, and we do not expect one person to have shipped all of this. We expect deep strength in the core, real interest in the frontier, and the judgment to choose the right tool per problem.
Core toolkit
- Language: Python (primary), with clean, production-grade code and API design.
- Large language models: LLM APIs and open models, prompt engineering, and retrieval-augmented generation (RAG).
- Retrieval and data: embeddings and vector databases (for example pgvector, FAISS, Chroma, or Pinecone), plus SQL and data-handling libraries such as pandas.
- Machine learning: ML and deep-learning libraries (for example PyTorch, scikit-learn, and the Hugging Face ecosystem).
- Serving and integration: deploying AI services behind REST APIs, containerized with Docker, with CI/CD, on a major cloud (AWS, GCP, or Azure).
The frontier we're building toward (beyond RAG)
- Agentic and tool-using systems: LLMs that call tools, take multi-step actions, and orchestrate workflows (function calling, agent frameworks). This is the main step beyond retrieval.
- Model adaptation: fine-tuning and parameter-efficient methods (for example LoRA/PEFT; SFT, and RLHF/DPO where useful), not only prompting.
- Multimodal for education: speech (ASR and TTS) for reading practice and audio answers, and vision/OCR/handwriting for scanned or drawn student work.
- Evaluation and experimentation as a discipline (the most valued senior skill): offline evaluation with golden datasets and LLM-as-judge, A/B testing, human-in-the-loop review, safety/guardrail evaluation, and tracing (for example Langfuse or OpenTelemetry).
- MLOps depth: experiment tracking (for example MLflow or Weights & Biases), model registry, serving and inference optimization, and monitoring for drift and quality.
- Learner modeling for adaptive learning: knowledge tracing and item response theory (IRT) to model what each student knows.
What we're looking for
- 5+ years of software engineering experience, with at least 2 spent building AI or machine-learning features that actually shipped to real users, not demos or research prototypes.
- Strong Python, and genuine breadth across modern AI: large language models, retrieval and embeddings, and machine-learning models, with the judgment to pick the right tool and the discipline to avoid AI for its own sake.
- The ability to model learner knowledge and progress, or the fundamentals and appetite to build that capability, so personalization is grounded in a real model of the student and not guesswork.
- A rigorous approach to evaluating and measuring output quality and learning impact, this matters more to us than fluency in any single framework.
- The ability to explain AI and adaptive decisions and trade-offs clearly to non-technical stakeholders, and to document your work so others can rely on it.
- Comfort working hand in hand with content and editorial teams, since adaptive learning is as much about learning design as it is about engineering.
- Solid machine-learning fundamentals and a clear sense of when a problem does or does not need a model.
- A responsible approach to data and privacy, especially where children and student records are involved.
Nice to have
- Familiarity with adaptive learning, knowledge tracing, item response theory (IRT), computer-adaptive testing, or spaced repetition.
- A background or genuine interest in learning science or psychometrics.
- Experience with self-hosted, open, or fine-tuned models.
- MLOps experience: deployment, monitoring, and cost control for AI services.
- Experience with Arabic language content or multilingual NLP.
- Experience in education, e-learning, or content-heavy platforms.
- Experience as the most senior AI person on a team or project.
- Inference serving and optimization at scale (for example vLLM, TGI, or Triton, plus quantization such as GPTQ or AWQ).
Working Hours:
- Monday to Friday: 8:00 AM to 5:00 PM including 1 hour lunch break
- Saturday: 8:00 AM to 1:30 PM
