You are the ranking lead for a travel marketplace with 6 million listings. Searches look like "beachfront villa in bali with a pool for 6 people in august": a mix of location, dates, party size, amenities and vague intent.
The current system is BM25 over listing text plus a hand-tuned boost for review score and booking count. It works acceptably for head queries and badly for the long tail, and it heavily favours established listings, so new hosts never get traffic.
The business metric is bookings, not clicks. A click that does not convert is close to worthless, and a booking that ends in a bad review is worse than no booking at all.
You have 18 months of search logs: queries, impressions with positions, clicks, and bookings.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
Your click logs say position 1 is always the best result. Why is that a problem, and how do you get unbiased training data?
Design the label. What exactly is a positive example, given that a booking with a 1-star review is worse than no booking?
A new host lists a beautiful property and gets no traffic for three months. Trace the loop and break it.
Minimum 8 components · needs a wide desktop screen