Dragon Candy
An AI agent that analyzes restaurant websites to draft campaign briefs and matches vetted creators in hours instead of weeks.
Case study by Talha Amir · AI Agent & Full-Stack Engineer
Discuss a similar project →- Dragon Candy · dragoncandy.io
- AI Full-Stack Engineer
- Ongoing
Overview
Dragon Candy is a two-sided content marketplace connecting hospitality brands with professional content creators. At its core is an autonomous AI workflow ('Donny') that extracts restaurant branding and menu data from any public URL, synthesizes creative shoot briefs with structured schemas, and matches high-affinity creators based on stylistic and audience scoring.
The Problem
Restaurants and hospitality brands require a consistent pipeline of high-quality photo and video content for social channels. However, traditional production involves cumbersome briefing documents, manual creator discovery, and lengthy negotiation cycles that take weeks — making reactive or seasonal promotions impossible.
- ▸Restaurant owners lack the time or creative terminology to draft detailed shoot briefs.
- ▸Manual creator matching was slow, subjective, and prone to mismatched aesthetics.
- ▸Coordination friction caused turnaround times to drag out for 2–3 weeks.
The Solution
Designed and built an end-to-end AI-powered marketplace platform. When a restaurant enters its website URL, the AI agent ingests publicly available content, identifies cuisine specialties and aesthetic tones, and outputs a structured production brief. The platform then computes matching scores against verified creators, coordinates bookings, and supports a rush 'DragonDash' turnaround lane.
- ✓Zero-friction onboarding: operators simply submit their URL.
- ✓Automated creative brief generation using strict JSON schema extraction.
- ✓Automated creator scoring and instant match recommendations.
- ✓Integrated Stripe payments, escrow holds, and creator payouts.
Architecture
End-to-end architecture bridging client dashboards, edge functions, structured LLM extraction, and transaction management.
AI & Brief Agent
OpenAI API with Structured Outputs (JSON Schema mode) extracting brand tone, color palette, cuisine focus, and shoot requirements from scraped URL content.
Frontend Layer
React, TypeScript, shadcn/ui, and Tailwind CSS powering responsive brand and creator portals with live status tracking.
Data & Auth Layer
Supabase PostgreSQL with Row Level Security (RLS), realtime subscriptions for chat and status transitions, and secure file storage.
Transaction Engine
Stripe Connect custom onboarding, automated escrow holds upon booking, and payout triggers on content approval.
Key Features
URL-to-Brief Agent
Extracts key dish categories, visual ambiance, and brand aesthetic from website markup into structured shot lists.
Intelligent Creator Matching
Scores creators by past portfolio style, proximity, audience demographics, and delivery turnaround reliability.
Dual-Sided Dashboards
Tailored operational views for restaurant managers (briefs, revisions, approvals) and creators (job feeds, upload portals).
DragonDash Express Delivery
Priority workflow allowing restaurants to book same-day shoots with guaranteed 24-hour turnaround.
Technology Stack
Full-Lifecycle AI Full-Stack Engineer
Owned application architecture, AI agent workflow implementation, database schema design, and production frontend engineering.
- ▪Architected the URL ingestion pipeline and structured output schemas for brief generation.
- ▪Implemented the creator-matching algorithm and Supabase database models.
- ▪Engineered both brand and creator web portals with responsive shadcn/ui components.
- ▪Integrated Stripe Connect checkout, payment holding, and automated payouts.
Product UI
Technical Challenges
Inconsistent Website Structures
Challenge: Target restaurant websites varied wildly in HTML structure, containing SPAs, social link trees, or PDF menus.
Solution: Implemented a resilient extraction pipeline that falls back to clean meta-tag synthesis and text chunking before passing to OpenAI's structured schema extractor.
Output Predictability & Format Integrity
Challenge: Shoot briefs needed strict field definitions (shot type, lighting style, deliverables) to be actionable for creators.
Solution: Enforced strict JSON Schemas with Zod parsing on the backend, ensuring zero hallucinated field names or broken contract payloads.
Outcome
Turned a multi-week manual production cycle into an automated workflow where restaurants receive verified briefs and matched creators within hours.
- ✓Deployed zero-configuration brief generator operating under 15 seconds per site URL.
- ✓Built and launched production two-sided marketplace on Supabase and React.
- ✓Automated the entire booking, payment escrow, and asset handoff loop.
- Hours
- Turnaround, not weeks
- Automated
- AI briefs from URL
- End-to-End
- Booking to payout
