TasteBuddy is a personal restaurant intelligence platform I built end-to-end. It lets users track every restaurant they visit, rank them across custom categories (Breakfast, Dinner, Tasting Menus, Ramen Night — whatever fits your life), manage a wishlist, and get AI-generated dining recommendations — all from one place.
The app solves a real fragmentation problem: most food-obsessed people scatter notes across Yelp, Google Maps, Beli, and saved Instagram posts. TasteBuddy consolidates everything into a structured, searchable record with social and AI layers on top.
Core features: visit logging (occasion, spend, party size, photos), a tiered ranking system (S/A/B/C across custom categories), Google Places–powered listing lookup, neighborhood Leaflet maps, an AI concierge, and a persona ecosystem for browsing other users' taste profiles.
TasteBoard is the personal analytics dashboard — visit streaks, cuisine breakdowns, spend trends, and crown awards (Most S-Tier Picks, Most Revisited, Widest Cuisine Range). The Tastie Score is a single engagement metric (0–1000) that grows as users rank more spots, log visits, explore cuisines, and follow personas — earning titles from Picky Eater to True Foodie.
Additional features: Photo Journal (photos tied to specific visits, permanently linked to rankings), Reservation Intel (save release dates + target dinner dates, get reminded when to book), and a Social Feed (public visit reactions and activity from followed users).
When a user asks the AI concierge a question (e.g. "Help me find a restaurant for our anniversary dinner — she loves Italian"), a two-stage agent pipeline fires. First, a question classifier decides whether the query needs live web data or can be answered from the local database alone — skipping Perplexity entirely for simple lookups (zero API cost). For richer queries, Perplexity sonar-pro runs two parallel searches: one across food media (The Infatuation, Eater, TimeOut, Beli, Resy, Instagram, TikTok, and more!) and one across Reddit threads — both steered by the user's S/A-tier taste profile. Claude Sonnet 4.6 then receives the combined research output and a structured snapshot of the user's actual taste history — their rankings, visit log, wishlist, and active persona — producing a recommendation personalized to that specific user, with citations and a TasteBuddy Connection explaining exactly why each pick fits their palate.