What "Great" Looks Like
Prompt & RAG Engineering
Design, tune, and A/B-test prompt chains that lift instruction-following accuracy, session-completion detection, and overall Buddy engagement. Implement RAG pipelines to power both support bots and in-product answers.
Multi-language Scale-out
Modularise Buddy prompt chains so every language pair uses a single, well-structured template with language-specific slots. Add capabilities to compare model performance for different languages.
Recommendations Engine
Prototype and productionise a Kotlin- or Python-based service that recommends the next best learning feature, measuring success via lift in feature completions and balanced usage.
Evaluation & Observability
Build an end-to-end prompt-quality harness (unit + offline + in-prod metrics) and integrate it with our custom model-serving layer.
Collaboration & Ownership
Partner daily with Sonia (Head of Content), Luis (Dir. Product Strategy & Innovation) and Elizaveta (Engineering Manager), proactively taking ownership of high-leverage deliverables that move the KPIs.
Requirements
- Must-Have
- Prompt-engineering best practices
- Proactiveness & strong ownership mindset
- Retrieval-Augmented Generation (RAG)
- Nice-to-Have
- Recommender-systems foundations
- A/B testing at scale
- Python (LLM tooling) & Kotlin fundamentals
- ML-ops tooling (Airflow, Feature Store)
- Basic statistics / experimentation design
