Tagline: Real food. Real fast. Real smart. Status: Concept / MVP (pre-seed) Last updated: 2026-06-05
Fast food is broken. The industry optimized for cost and speed at the expense of quality, transparency, and health. Meanwhile, premium fast-casual (Chipotle, Sweetgreen) proved people will pay more for real food — but the kitchen model hasn't changed in decades.
Parch'd rethinks the fast-casual kitchen from first principles. We take the French culinary technique of en papillote (parchment-paper steaming) — a 200-year-old method prized for locking in flavor and moisture — and combine it with a Subway-style assembly line, all orchestrated by AI.
The result: a 3-5 minute meal where every ingredient is visible through the paper, the aroma is the marketing, and the AI makes every decision smarter than a human ever could.
This is not a restaurant that uses AI. This is an AI-native restaurant — AI is the operating system, not a feature.
[Ingredients arrive] → [CV quality check] → [AI predicts demand: prep schedule]
↓
[Customer order] → [AI oven controller: custom temp curve per combo]
↓
[3-5 min bake] → [Seal integrity check] → [Serve]
Components:
- Computer Vision Quality Control: Cameras at receiving dock grade produce freshness. Rejects sub-par ingredients before they enter inventory.
- Predictive Prep: ML models forecast demand per ingredient per 15-min window. Pre-chopped vegetables, pre-portioned proteins. Near-zero food waste.
- AI Oven Controller: Each protein-vegetable-sauce combination has an optimal temperature curve. The oven reads the order's barcode and auto-adjusts. No chef guesswork.
- Smart Sealer: Automated parchment folding and sealing. Consistent seal = consistent steam = consistent quality.
[Customer enters] → [Face/fob recognition: load profile]
↓
[AI Nutritionist: "Based on your goals, try salmon + asparagus + citrus"]
↓
[Customize at screen or voice: "Swap asparagus for broccolini"]
↓
[Dynamic pricing display: real-time price reflecting demand + ingredient cost]
↓
[Pay → Watch your pouch get sealed → 3 min → aromatic reveal]
Components:
- AI Nutritionist: Multi-modal LLM that knows your health profile, dietary restrictions, fitness goals, and taste history. Recommends combinations you'll actually like.
- Flavor Profile Learning: Collaborative filtering across all customers. "People who liked your last 3 meals also enjoy..."
- Natural Language Ordering: Voice or text. "I want something light, under 400 calories, with white fish." No menu navigation needed.
- Dynamic Digital Menu Boards: Prices adjust in real-time based on ingredient freshness (discount near-expiry), demand spikes, and time of day.
[All stores] → [Central data lake] → [ML models]
↓
┌─────────────────────┼─────────────────────┐
↓ ↓ ↓
Supply Chain Dynamic Pricing Location Intel
(predict → order (revenue-maximizing (where to open
→ minimize waste) per store/hour) next + when)
Components:
- Supply Chain Brain: Predicts ingredient demand 7 days out per store. Auto-orders from suppliers. Tracks actual vs predicted to continuously improve.
- Dynamic Pricing Engine: Revenue management like airlines/hotels. Lunch rush premium, late-afternoon discount. A/B tests pricing strategies across stores.
- Location Intelligence: ML model trained on successful store features (foot traffic, office density, income demographics, competitor proximity). Scores candidate locations.
- Staff Scheduling Optimizer: Matches predicted demand curves to optimal staffing. Cuts labor costs 15-20% vs fixed schedules.
- Real-Time P&L Dashboard: Every store is a live data point. COGS, labor, throughput, waste — all tracked to the minute.
| Element | Direction |
|---|---|
| Name | Parch'd (or localized: 纸包鲜 / Parch'd 纸包鲜 / ...) |
| Visual | Clean white space + warm kraft paper texture + vibrant ingredient color pops |
| Typography | Modern sans-serif with a hint of editorial elegance |
| Photography | Overhead shots of parchment pouches mid-open, steam rising, ingredients visible |
| Architecture | Open kitchen as theater. Oven wall as the "stage." Customers watch their pouch go in, come out. |
| Vibe | French bistro efficiency × Silicon Valley transparency × Japanese attention to detail |
| Core Promise | See every ingredient. Taste every flavor. Zero mystery. |
The parchment pouch is not just a cooking method — it's the brand's physical manifestation:
- Transparency: You see your food before it's cooked. No hidden fryer, no mystery griddle.
- Aroma marketing: When 20 pouches open simultaneously, the whole street smells like roasted garlic and herbs.
- Instagram-native: The pouch opening is a ritual. Steam + aroma + reveal = highly shareable moment.
- Health signaling: Paper = clean, natural, no oil bath. The packaging IS the message.
- Primary: Urban professionals, 25-40, health-conscious, time-poor, willing to pay ¥40-60 for lunch
- Secondary: Fitness enthusiasts (post-workout protein + vegetables), flexitarians
- Tertiary: Families looking for a "real food" alternative to McDonald's
| McDonald's | Subway | Wagas/Gaga | Parch'd | |
|---|---|---|---|---|
| Price | ¥25-35 | ¥30-40 | ¥50-80 | ¥40-60 |
| Speed | 2 min | 3 min | 8-12 min | 3-5 min |
| Health | Low | Medium | High | High+ (visible) |
| Transparency | None | Assembly line | Kitchen hidden | Stage kitchen |
| Tech | Kiosk | Kiosk | QR order | AI-native full stack |
| Metric | Estimate |
|---|---|
| Avg ticket | ¥48 |
| Daily covers (lunch) | 200-300 |
| Daily covers (dinner) | 100-150 |
| Monthly revenue | ¥430K - ¥650K |
| COGS (target) | 28-32% |
| Labor (AI-optimized) | 18-22% (vs 25-30% industry) |
| Rent | 12-15% |
| Store-level EBITDA | 25-30% |
| Payback period | 12-18 months |
- Core: Per-meal sales (85% of revenue)
- Subscription: Weekly meal plans with AI-curated menus, 15% discount (10%)
- Data Licensing: Anonymized nutrition insights for health/insurance partners (5%, mid-term)
- Franchise: AI-powered franchise-as-a-service (long-term)
This repo delivers three artifacts:
→ This document (PLAN.md) + bilingual READMEs
src/papillote/
├── models.py # Data models: Menu, Order, Customer, Store
├── engine.py # AI engines: demand prediction, pricing, recommendations
├── simulator.py # Discrete-event restaurant simulator
Key capabilities:
- Demand forecasting: Time-series prediction of item demand per 15-min window
- Dynamic menu pricing: Elasticity-based pricing with freshness discounting
- Customer personalization: Content-based + collaborative filtering recommendation
- Full-day simulation: Customer arrival → order → kitchen queue → serve → revenue report
app/
└── streamlit_app.py
A Streamlit web app that serves as both brand landing page and interactive prototype:
- Brand story with visual identity
- Interactive menu builder (customize your papillote)
- Live restaurant simulator dashboard
- AI nutritionist demo
- Unit economics calculator
┌─────────────────────────────────────────────────────────┐
│ CLOUD LAYER │
│ ┌─────────────┐ ┌──────────┐ ┌───────────────────┐ │
│ │ Demand Brain │ │ Pricing │ │ Location Intel │ │
│ │ (Time-series │ │ Engine │ │ (Geospatial ML) │ │
│ │ Transformer)│ │ (RL) │ │ │ │
│ └──────┬───────┘ └────┬─────┘ └────────┬──────────┘ │
│ └────────────────┼────────────────┘ │
│ ┌──────┴──────┐ │
│ │ Data Lake │ │
│ └──────┬──────┘ │
└──────────────────────────┼──────────────────────────────┘
│
┌──────────────────────────┼──────────────────────────────┐
│ EDGE (per store) │
│ ┌──────────────┐ ┌────┴─────┐ ┌──────────────────┐ │
│ │ CV Quality │ │ Oven │ │ Customer Display │ │
│ │ (Jetson Nano)│ │ Controller│ │ (AI Nutritionist) │ │
│ └──────────────┘ └──────────┘ └──────────────────┘ │
│ │
│ ┌──────────────────────────────────────────────────┐ │
│ │ Store Gateway (local sync) │ │
│ └──────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────┘
- Edge: NVIDIA Jetson for CV + oven control. Works offline if cloud is down.
- Cloud: Central ML training, demand forecasting, menu optimization.
- App: Customer-facing mobile app (RN/Flutter) with AI nutritionist chat.
- IoT: Smart ovens with REST API for temperature curve control.
| Phase | Timeline | Milestone |
|---|---|---|
| Phase 0: Concept | Now | This repo — plan, prototype, pitch |
| Phase 1: Pilot | Months 1-6 | Single store. Full AI stack. Prove unit economics. |
| Phase 2: Cluster | Months 7-12 | 3 stores in one city. Prove replication + supply chain AI. |
| Phase 3: Series A | Month 12-18 | Raise on data: unit economics + AI moat. |
| Phase 4: City | Month 18-24 | 15+ stores. Franchise model launch. |
| Phase 5: Multi-City | Year 3+ | New markets. AI fully autonomous operations. |
| Risk | Severity | Mitigation |
|---|---|---|
| Paper pouch = slow throughput | High | 3-5 min target validated. Batch oven design. Pre-assembly line. |
| Chinese market prefers cooked/hot food | Medium | "Paper-baked" = hot, aromatic. Not cold salad. Aligns with Chinese preference. |
| AI over-automation → sterile experience | Medium | Open kitchen as theater. Human interaction at reveal moment. |
| Supply chain for parchment + fresh fish | Medium | Partner with established seafood distributors. Paper is commodity. |
| Copycats (no IP moat) | High | AI system is the moat. 3-layer data flywheel. Brand + speed. |
- AI maturity: Computer vision, demand forecasting, and LLMs are production-ready and cheap.
- Health consciousness: Post-COVID, Chinese consumers are spending more on quality food.
- Labor costs rising: AI kitchen automation directly addresses China's rising labor costs.
- Delivery fatigue: People want to eat OUT again — but want better experiences than 2019.
- No incumbent: No one has built an AI-native restaurant chain from scratch. First-mover advantage in a trillion-yuan market.
- Chipotle: Proved assembly-line customization scales
- Sweetgreen: Proved premium fast-casual health works
- Haidilao: Proved service theater is a moat in China
- Amazon Go: Proved AI-native retail is possible
- Tesla: Proved building the OS + the hardware together wins