NutriPrompt is an AI decision-support architecture designed to transform nutrition planning into a contextual, validated and explainable workflow.
NutriPrompt is not a meal-plan generator.
It is a production-oriented AI system built to demonstrate how modern GenAI products should be designed:
- context-aware
- retrieval-augmented
- validation-first
- resilient
- explainable
- execution-ready
Built with Django, Python, RAG, OCR, multi-provider LLM orchestration, validation engines, PDF generation, and structured output pipelines.
Its purpose is simple:
to show how AI moves from content generation into intelligent product systems.
👉 https://nutriprompt-live.streamlit.app
This public Streamlit demo exposes NutriPrompt’s internal AI workflow in a recruiter-friendly and technical format:
- structured intake
- RAG orchestration
- OCR processing
- validation layers
- AI Copilot explainability
- structured outputs
NutriPrompt is an AI Systems Engineering project.
It demonstrates:
- Prompt Engineering for controlled generation
- Retrieval-Augmented Generation (RAG)
- OCR ingestion and structured parsing
- Validation layers before inference
- Multi-provider LLM orchestration
- Fallback resilience strategies
- Explainable outputs
- Human-readable + machine-readable reports
- Product-oriented AI UX design
- Context-first AI workflows
This reflects how real-world AI systems move beyond prompting into structured, reliable and explainable pipelines.
NutriPrompt demonstrates how AI can:
- reduce planning friction
- improve recommendation consistency
- validate restrictions before execution
- increase personalization quality
- reduce user decision fatigue
- transform static advice into actionable workflows
This is not just about nutrition.
It is about building AI products that solve operational problems reliably.
NutriPrompt’s architecture is transferable.
The same workflow pattern can be applied to:
- healthcare decision-support
- legal document analysis
- HR screening workflows
- customer support copilots
- insurance validation systems
- procurement workflows
- industrial compliance systems
This makes NutriPrompt a reusable AI systems pattern.
Not a niche app.
Nutrition planning is not a content generation problem.
Reliable AI nutrition systems must understand:
- personal goals
- digestive conditions
- ingredient compatibility
- dietary restrictions
- food preferences
- budget limitations
- real-life constraints
- execution friction
NutriPrompt treats nutrition as an intelligent workflow problem.
Not as a chatbot.
That distinction changes everything.
Its architecture enriches every request before generation.
This produces:
✅ grounded outputs ✅ hallucination-resistant pipelines ✅ explainable decision flows ✅ compatibility-aware planning ✅ resilient multi-provider orchestration ✅ production-grade AI workflows
I built NutriPrompt to explore how AI systems evolve beyond content generation into decision-support architectures.
The goal was never to generate meal plans.
The goal was to design a system that demonstrates:
- structured reasoning
- domain grounding
- retrieval orchestration
- explainability
- resilient fallbacks
- product-oriented thinking
- real-world execution
This project reflects how I approach AI product design:
connecting business, data and technology into usable systems.
User Input
↓
Structured Intake Forms
↓
Profile Analysis
↓
RAG Knowledge Retrieval
↓
Prompt Builder
↓
Gemini API
↓ (Fallback)
OpenAI API
↓
Structured JSON Output
↓
Nutrition Rules Engine
↓
Validation Layer
↓
HTML Rendering
↓
Shopping Intelligence
↓
PDF Generation
↓
AI Copilot Layer
Transforms structured user input into contextual AI workflows:
- goals
- digestive symptoms
- restrictions
- food preferences
- budget logic
- lifestyle context
Produces:
- weekly meal plans
- compatibility-aware recommendations
- tupper adaptation
- explainable profile tags
- downloadable PDF
Transforms recommendations into execution:
- categorized shopping lists
- grouped ingredients
- optimized planning
- action-ready outputs
- Personalized Nutrition Planning
- Retrieval-Augmented Generation (RAG)
- OCR + Ingredient Intelligence
- Validation Engine
- Shopping Intelligence
- AI Copilot Explainability
- Multi-provider resilience
- Structured outputs
Gemini API
↓
OpenAI Fallback
↓
Structured Mock Generation
Benefits:
- stable demos
- graceful degradation
- provider independence
- predictable outputs
NutriPrompt was intentionally designed around:
- Controlled generation over open prompting
- Retrieval before inference
- Validation before recommendation
- Explainability over black-box generation
- Provider abstraction
- Structured outputs
- Service-oriented architecture
- Product-first UX thinking
These principles reflect modern AI product architecture.
| Layer | Technology |
|---|---|
| Backend | Django |
| Language | Python 3.13 |
| AI Providers | Gemini API + OpenAI API |
| Retrieval | Custom Nutrition RAG |
| OCR | Tesseract OCR |
| Data | JSON |
| PDF Rendering | WeasyPrint |
| Frontend | HTML + CSS |
| Demo Layer | Streamlit |
| Testing | Django Test Framework |
Current validation includes:
- prompt generation
- knowledge retrieval
- validation logic
- OCR processing
- context injection
- structured outputs
- shopping generation
python manage.py test17 automated tests passing
git clone https://github.com/beatriangu/NutriPrompt.git
cd NutriPrompt
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txtRun Django:
python manage.py runserverRun Streamlit demo:
streamlit run streamlit_demo/app.pyNutriPrompt provides informational guidance only.
It does not replace medical, nutritional or healthcare advice.
Outputs should be reviewed by qualified professionals.
Bea Lamiquiz
🌐 Portfolio: https://bchill.net 💻 GitHub: https://github.com/beatriangu 💼 LinkedIn: https://www.linkedin.com/in/bealamiquiz/
I build AI systems at the intersection of:
- Product
- Data
- Business Logic
- Generative AI
Open to:
- AI Product roles
- Applied AI opportunities
- Technical collaborations
- AI architecture discussions
- Product strategy conversations