About the challenge

You'll be handed a messy, real-world-style software product catalogue. Your job: whip it into shape according to our data guidelines, ingest it, and then build a recommendation engine smart enough to ask the right questions, zero in on what a customer actually needs, and justify its top 3 picks like a human analyst would.

This isn't a toy problem; it's a compressed version of a real challenge our product team deals with regularly. We want to see how you think, not just what you can copy-paste.

• Solve a real problem, not a puzzle. This mirrors an actual challenge our product team works on — your solution isn't thrown away at 5pm.
• Get hands-on with LLM-powered engineering. Build with a live local LLM key — embeddings, reasoning, conversational probing, your call on approach.
• Fast-track your interview. Standout performers get priority fast-tracked interviews with the hiring team — this is as much an audition as it is a hackathon.
• Walk away with a portfolio piece. A working recommendation engine with reasoning output is a strong project to show off, wherever you land next.
• Mentors on the floor. Our engineers will be around throughout the day if you get stuck or want a sanity check — not just judges who show up at the end.
• Good food, good people. Pizza, wifi, and a room full of people who like solving hard problems.

• You'll receive a sample software product catalogue (intentionally imperfect).
• Clean, normalize, and validate it against our product data guideline (schema, required fields, formatting rules — provided at kickoff).
• Ingest the sanitized dataset into a structure your recommendation engine can query.

• Build an engine that takes a customer query (e.g., "I need a CRM for a 20-person sales team that integrates with Slack") and returns the top 3 product recommendations.
• Your engine must include a probing layer — it should ask clarifying questions when the query is vague or under-specified, rather than guessing blindly.
• For each of the top 3 results, output:
  • Why this product was recommended (reasoning tied to the actual query/probing answers)
  • What would make it a better fit — i.e., what's missing, or what could improve the match

• Data quality
  How well the sanitized dataset conforms to the guideline, and how edge cases were handled
• Recommendation logic
  Relevance and accuracy of the top 3 suggestions
• Probing quality
  Are the clarifying questions genuinely useful, or just noise?
• Reasoning
  Clarity and specificity of the "why" and "how to improve" explanations
• Engineering craft
  Code quality, structure, and use of the LLM/tools available