Edible Dialect
iOS restaurant discovery app · App Store, Aug 2026 · Solo, under Overview Ventures LLC
A login-free NYC restaurant discovery app: structured quick search with fuzzy matching, badge-driven browsing (price tier, Michelin, James Beard), and Edible Experiences, a Claude-powered AI concierge. Under it all, a 100-point Edible Score aggregating multiple rating sources across 26,000+ restaurants.
I built Edible Dialect by directing a two-agent AI development workflow — defining the architecture, setting up the repos and project structure, and reviewing and integrating each agent's output, with handoffs at the API contract. The system-level calls were mine: agentic tool-calling for the concierge over single-shot RAG (and direct tool use over MCP); anonymous device identity as the foundation for monetization and per-device rate limiting — no accounts, no login friction; and the v2→v3 scoring redesign, moving from percentile normalization to user-intuitive absolute scores.
OpenClaw agent — backend
Executed the FastAPI service, data pipeline, and ops end to end: REST endpoints, search architecture (borough/cuisine/neighborhood decomposition, pg_trgm fuzzy matching), Edible Score implementation and full-corpus recomputes, concierge candidate retrieval and rate limiting, infra, and post-launch data operations.
Python 3.12 · FastAPI · asyncpg · Pydantic v2 · PostgreSQL 16 · Cloudflare R2 · NYC Open Data + Google Places ingest · YOLO + LLaVA photo audit
Claude Code — iOS
Executed the entire client: all SwiftUI screens, components, and animations; design system and badge execution; concierge integration; App Store Connect and TestFlight submission.
SwiftUI · MapKit · CoreLocation · StoreKit 2 · RevenueCat
Beyond the build itself, I set the product vision, all UX and design direction, data source selection and curation, and the App Review strategy — two rejections overturned via rebuttal of the same build.