SaaS platform for professional kitchen management — AI-powered recipe ingestion, cost analysis, and multi-establishment operations. Built by a professional chef with 8 years in the kitchen and 28 years in IT.
🔗 chefsoffice.com.br — free tier available
Showcase repository — full source available on request for technical evaluation.
Professional chefs manage dozens of technical recipe sheets scattered across PDFs, spreadsheets, and paper notebooks. Costing a new dish means manually entering ingredients, applying correction factors, and recalculating every time a supplier changes a price. Scaling from one kitchen to multiple establishments multiplies the chaos.
Cloud tools exist but they're either too generic (built by developers who never cooked) or too expensive for independent restaurants and small chains.
Chefs Office was built from the inside out — by someone who actually worked the line for 8 years.
- AI-powered recipe ingestion — upload a PDF, Word doc, spreadsheet, or photo of a handwritten recipe and Gemini 2.0 Flash extracts ingredients, quantities, preparation steps, and phases automatically
- Smart ingredient resolution — 4-layer matching: user catalog → 320-item global catalog (TACO/USDA) → sub-recipes → inline creation. No manual lookup.
- Automatic cost calculation — CMV%, price per portion, markup by factor, margin analysis — recalculated in real time as ingredient prices change
- Chained sub-recipes — an intermediate sheet (FTI) can be used as an ingredient inside a final sheet (FTP), with automatic cost propagation through the chain
- Professional PDF exports — recipe sheet, cost analysis (gerencial), and ANVISA-compliant nutritional report (RDC 429/2020), all generated in one click
- Multi-establishment — one account, multiple kitchens, independent pricing per location
- Nutritional data — calories, protein, carbs, fat, fiber from the TACO table (UNICAMP/USDA) with correction factors (FC/FCC) per category
The AI import zone is always visible at the top — drag and drop any PDF, DOCX, XLSX, or image to start.
Each ingredient shows FC (correction factor), calculated gross quantity, and cost in real time. Sub-recipes link inline as "Ficha Intermediária vinculada".
Steps organized by phase: pre-prep → prep → finishing → plating. Each step is editable with optional timing.
The financial panel at the bottom of every recipe: total cost, markup factor input, CMV%, suggested sale price, and profit per portion — all live. This is what generates the Gerencial PDF.
The SmartImporter resolves each ingredient across 4 layers simultaneously, showing similarity scores ("Seu catálogo · 40%", "TACO", "Ficha Intermediária"). Items with no match are created inline.
320-item TACO/USDA catalog browseable by category. Any item can be imported to the establishment's catalog with one click, then priced per supplier.
Menus organized by category: Entrada, Prato Principal, Acompanhamento, Porções, Tira-gosto, Drink, Sobremesa, Prato do Dia.
One account manages multiple establishments with independent ingredient pricing per location. In active use for professional culinary consulting.
Three report types generated from every recipe — see sample files in docs/samples/:
| Report | File | Description |
|---|---|---|
| Recipe Sheet | sample-receita-carpaccio.pdf | Photo, ingredients with FC/FCC, preparation steps by phase. Goes to the kitchen. |
| Cost Analysis | sample-gerencial-carpaccio.pdf | Cost per ingredient, CMV%, markup, profit, suggested price. Marked CONFIDENTIAL. Stays in the owner's drawer. |
| Nutritional Report | sample-nutricional-carpaccio.pdf | ANVISA RDC 429/2020 compliant. Per-portion and per-100g breakdown + per-ingredient contribution table. |
| Layer | Technology |
|---|---|
| Frontend | React 18 + TypeScript + Vite + Tailwind CSS + shadcn/ui |
| State management | TanStack Query v5 |
| Backend / Database | Supabase (PostgreSQL + Row Level Security + Auth + Storage) |
| AI pipeline | Gemini 2.0 Flash via Supabase Edge Function (Deno) |
| PDF generation | jsPDF + html2canvas |
| Automation | n8n (self-hosted, zero cost) |
| Deploy / CI | Lovable |
Infrastructure cost in production: R$110/month ($22 USD)
Browser (React 18 + TypeScript)
│
├── Supabase Auth (JWT) — email/password + Google OAuth
├── Supabase PostgreSQL — RLS on every table
├── Supabase Storage — recipe photos
└── Edge Function (Deno) — AI processing, no API key on client
│
├── POST /process-recipe-etl
│ ├── Receives: PDF / DOCX / XLSX / image
│ ├── Extracts text (DOCX/XLSX via XML unzip)
│ ├── Calls Gemini 2.0 Flash with fallback chain
│ │ gemini-2.0-flash → gemini-1.5-flash-8b → gemini-flash-latest
│ └── Returns: structured ExtractedRecipe JSON
│
└── POST /delete-user
profiles
establishments ──── establishment_members
└── establishment_ingredients ──── ingredients (global catalog: user_id IS NULL)
└── ingredients (user catalog: user_id = X)
recipes ──── recipe_ingredients ──── ingredients
└── preparation_steps └── sub_recipe_id → recipes (chained FTI)
menus ──── menu_sections ──── menu_recipes ──── recipes
Key design decisions:
- Global ingredient catalog (
user_id IS NULL) is read-only — imported via RPC before use pg_trgmsimilarity search for smart ingredient matching during AI import- Database constraint prevents
recipe_ingredientsfrom referencing ingredients without an owner - All AI processing happens server-side in Deno Edge Functions — Gemini API key never reaches the browser
User uploads file (PDF / DOCX / XLSX / PNG / JPG)
│
▼ Edge Function detects file type
│ ├── DOCX/XLSX → unzip XML, extract text
│ └── PDF/image → base64 encode (Gemini inline_data)
│
▼ Gemini 2.0 Flash extracts structured data
│ └── Fallback: gemini-1.5-flash-8b → gemini-flash-latest
│
▼ SmartImporter resolves each ingredient in 4 layers
│ 1. User's catalog (exact + pg_trgm similarity match)
│ 2. Global TACO/USDA catalog (320 ingredients)
│ 3. Existing sub-recipes (FTI)
│ 4. Inline creation if no match found
│
▼ AI-inferred preparation steps marked with visual badge
→ User reviews and confirms before saving
Built with a deliberately structured AI-first workflow:
| Tool | Role |
|---|---|
| Claude Code | Complex logic, SQL, RPCs, architecture decisions |
| Gemini | Frontend components, UI iteration |
| Lovable | React/Tailwind UI generation, CI/CD pipeline |
| n8n (self-hosted) | Notification automation, zero operational cost |
| Feature | Status |
|---|---|
| Core recipe management | ✅ Production |
| AI recipe import (PDF/DOCX/XLSX/image) | ✅ Production |
| Multi-establishment | ✅ Production |
| PDF export — recipe + cost analysis + nutritional | ✅ Production |
| ANVISA RDC 429/2020 nutritional report | ✅ Production |
| 320-item TACO/USDA ingredient catalog | ✅ Production |
| Google OAuth | ✅ Production |
| Pro / billing system | 🔄 In development |
| Mobile app | 📋 Planned |
| Supplier price integration | 📋 Planned |
Built by Conrado Nogueira — 8 years as a professional chef + 28 years in IT infrastructure.
The domain knowledge in this product is not cosmetic. The correction factor system, the chained sub-recipe architecture, the 4-phase preparation step model (pre-prep → prep → finishing → plating), the "USO INTERNO — CONFIDENCIAL" stamp on the cost analysis — these details come from real kitchen experience, not assumptions.
In active use for professional culinary consulting.
github.com/JConradoN · Available for freelance projects (USD/EUR)
Showcase repository — full source available on request for technical evaluation.