A staff-level take-home for AI engineering candidates: build a knowledge graph from exercise data and clinical ontologies, then a coach dashboard on top of it — an AI workout generator and a member-context copilot — that gives safe, personalized, explainable recommendations.
- Time: 1 day
- Stack: your choice — we want you to pick the tools and defend them
- Data: synthetic only (provided in
data/); never use real member data
| Path | Purpose |
|---|---|
ASSESSMENT.md |
The full take-home spec — task, knowledge graphs, ontologies, build steps, deliverable |
data/exercises.json |
Exercise catalog (50 exercises) |
data/member-context.json |
One rich synthetic member: profile, goals, injuries, chat history, biomarkers, labs (blood panel + DEXA), adherence, churn signals |
Two surfaces in one coach dashboard:
- Workout Generator — a prompt + time form that calls an agentic runtime and renders a structured workout. It reasons over a movement/clinical knowledge graph (grounded in ontologies like OPE, COPPER, SNOMED CT, PROV-O, SKOS) to keep recommendations injury-aware, equipment-aware, and explainable.
- AI Copilot — a chat panel with retrieval over a member-context knowledge graph: adherence trends, sleep, churn risk, the morning brief, charts, and past conversations.
See ASSESSMENT.md for the complete spec.
Build in a GitHub repo with a comprehensive README (architecture and tech-choice rationale, how to run locally, how you used AI, and your trade-offs). Use synthetic data only, then share the link.