Gift Matchr

Engages users in a multi-turn dialogue to progressively gather recipient context (age, interests, relationship, occasion, budget) through natural language questions rather than forms. Uses turn-by-turn conversation state management to build a mental model of the gift-giving scenario, with each response informing subsequent clarifying questions. The system maintains conversation history to avoid redundant questions and refine understanding based on user corrections or elaborations.

Synthesizes gathered context (budget, age, interests, occasion, relationship type, recipient personality) into ranked gift suggestions by prompting an LLM to generate ideas that balance multiple competing constraints. The system likely uses prompt engineering to weight criteria (e.g., 'budget is hard constraint, interests are soft constraint') and generate 3-7 diverse suggestions rather than a single recommendation. Each suggestion includes a brief rationale explaining why it matches the recipient profile.

Filters and contextualizes gift suggestions based on the specific occasion (birthday, holiday, wedding, thank-you, apology) and relationship type (friend, family, colleague, acquaintance, romantic partner) to avoid socially inappropriate recommendations. The system applies implicit rules or learned patterns (e.g., 'romantic gifts for spouses differ from gifts for colleagues') to weight suggestions and exclude categories that don't fit the context. This filtering happens during recommendation synthesis, not as a post-processing step.

Generates gift suggestions that respect hard budget constraints by incorporating price ranges into the LLM prompt and filtering suggestions to fall within the specified budget. The system likely uses estimated price ranges for common gift categories (e.g., 'luxury watches: $200-500', 'books: $10-30') to guide generation. Suggestions may include price estimates, though these are not verified against real-time retail data. The system can handle budget ranges (e.g., '$50-100') and may suggest combinations of smaller items if a single item exceeds budget.

Tailors gift suggestions to the recipient's stated interests and hobbies by extracting key themes from the conversation (e.g., 'photography', 'cooking', 'gaming', 'reading') and using them to guide recommendation generation. The system maps broad interest categories to specific gift ideas (e.g., 'photography' → camera accessories, photo books, lighting equipment) and prioritizes suggestions that align with these interests. This personalization is implicit in the LLM prompt rather than explicit category matching.

Filters and contextualizes gift suggestions based on the recipient's age to ensure developmental appropriateness and safety. The system applies implicit age-based rules (e.g., 'no small choking hazards for toddlers', 'age-appropriate content for children', 'mature interests for adults') during recommendation generation. Age ranges are likely mapped to broad categories (toddler, child, teen, young adult, adult, senior) with different gift profiles for each. The system may also consider age-related interests (e.g., 'teens prefer tech and fashion' vs. 'seniors prefer comfort and nostalgia').

Maintains conversation state across multiple turns within a single session, tracking gathered context (recipient profile, budget, occasion, interests) and using it to avoid redundant questions and provide coherent follow-ups. The system stores conversation history in client-side or server-side state (likely session storage or temporary backend cache) and uses it to inform subsequent LLM prompts. State is reset on new conversation or page reload, with no persistent cross-session memory. The system may use conversation context to refine recommendations if the user provides feedback or corrections.