+ ]Qj#BRt^RItRRltRRltRRltRR ltR#) aSystem prompt builders for each endpoint. Kept separate from app/services/ so the prompt text (the thing most likely to need tuning against real model output) can change without touching request/response wiring, and separate from app/schemas.py so the JSON shape described to the model and the JSON shape enforced by pydantic can be eyeballed side by side. Every prompt below follows the same contract established in goplaces_backend/v4/main.py's _get_itinerary_prompt(): describe the exact JSON shape, demand JSON-only output with no markdown fences, and provide the fallback parser in providers/json_utils.py for when a model ignores that instruction anyway. Nc V^8dQhR\R,R\R,R\\,R\R\/#) trip_specN preferencesanswered_questions num_optionsreturndictlistintstr)formats"3C:\xampp\htdocs\meridian\meridian-ai\app\prompts.py __annotate__rsM\ \ d{\ \ T \  \  \ cRT RT R\P!T;'g/^R7 R\P!T;'g/^R7 R\P!T;'g.^R7 R2 P4#)z You are Meridian's travel itinerary generation engine. Given a trip brief, traveler preferences, and (optionally) real flight/stay/event search candidates, produce a distinct day-by-day itinerary options. Respond with STRICT JSON only - no markdown fences, no commentary before or after the object - matching exactly this shape: { "options": [ { "letter": "A", "name": "", "is_ai_pick": true, "summary": "<1-2 sentence pitch explaining why this option fits>", "total_cost": {"amount": 2400, "currency": "USD", "is_estimate": true}, "flights": [ { "airline": "", "flight_number": "", "departure_airport": "", "arrival_airport": "", "departure_datetime": "", "arrival_datetime": "", "cost": 450, "currency": "USD", "booking_url": "", "is_estimate": false } ], "stays": [ { "name": "", "address": "", "check_in_date": "", "check_out_date": "", "room_type": "", "cost_per_night": 180, "currency": "USD", "booking_url": "", "is_estimate": false } ], "days": { "1": { "title": "", "destinations": { "": { "activities": ["", ""], "estimated_cost": {"amount": 80, "currency": "USD", "is_estimate": true} } } } } } ] } Rules: - Generate exactly u options: one clearly the best overall fit for the stated preferences (mark "is_ai_pick": true on that one only), one value-oriented, one premium/wildcard alternative. - Every day of the trip must appear as a key in "days" - never skip a day. - FLIGHTS: If flight_candidates are supplied, choose the best matching flight(s) from the list and populate the "flights" array. Copy real prices (is_estimate: false) and include the candidate's booking_url if it has one. For multi-city or connecting trips add one entry per leg. If no candidates, still include a plausible flight with estimated pricing (is_estimate: true) whenever the trip requires flying. - STAYS: If stay_candidates are supplied, choose the best fit hotel for this option and populate the "stays" array. Use the candidate's price as cost_per_night and its link field as booking_url. Different options (A/B/C) should pick different candidates where possible. If no candidates, still include a plausible hotel with estimated pricing. - EVENTS: If event_candidates are supplied, weave the most relevant real events into the appropriate days under "destinations" — use the event name as the destination key and list event details in "activities". Do not leave event_candidates unused. - "days" keys are day numbers as strings ("1", "2", ...), not dates. - Do NOT include flight or hotel info inside "days" — those belong in "flights"/"stays". TRIP SPECIFICATION: indentz TRAVELER PREFERENCES: z! ANSWERED CLARIFYING QUESTIONS:  jsondumpsstrip)rrrrs&&&&ritinerary_system_promptrs  8p =!(IOOA&'(K  2a()*  $ $"Q/01kV l EGmV rcXV^8dQhR\R,R\R,R\/#)rrNrr)r r )rs"rrrps*  td{  QT rcR\P!T;'g/^R7 R\P!T;'g/^R7 R2P4#)ay You are Meridian's pre-generation clarifying-question assistant. Your job is to spot ambiguity in a trip brief BEFORE an itinerary is generated from it - missing budget, unclear travel style, vague or conflicting preferences - and ask the smallest set of questions that would most improve generation quality. Respond with STRICT JSON only, matching exactly this shape: { "questions": [ { "id": "q_01", "text": "", "type": "single_choice", "options": ["