NutriPrompt is a domain-specific AI workflow system designed to demonstrate production-grade Prompt Engineering, Retrieval-Augmented Generation (RAG), OCR pipelines, validation layers and resilient multi-provider LLM orchestration.

Built with Django, Prompt Engineering, Retrieval-Augmented Generation (RAG), OCR pipelines, validation layers, shopping intelligence, PDF generation, and multi-provider AI orchestration.

NutriPrompt is not a meal-plan generator.

It is an AI decision-support system for contextual nutrition planning.

Its purpose is to demonstrate how modern AI systems combine:

• structured user data
• domain knowledge
• retrieval systems
• validation layers
• explainability
• fallback orchestration
• execution-ready outputs

to deliver more reliable, grounded and operational AI workflows.

👉 https://nutriprompt-live.streamlit.app

This public technical demo exposes NutriPrompt’s internal workflow in a recruiter-friendly format:

• structured intake
• RAG orchestration
• OCR reasoning
• validation layers
• AI Copilot explainability
• structured outputs

NutriPrompt is designed as an AI systems engineering project.

It demonstrates:

• Context-first AI workflows
• 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

This project reflects how production AI systems move beyond raw prompting into structured, resilient and explainable workflows.

Nutrition planning is not a simple 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.

Its architecture enriches every request before generation.

This produces:

✅ grounded outputs
✅ fewer hallucinations
✅ explainable recommendations
✅ compatibility-aware planning
✅ resilient provider orchestration
✅ production-oriented AI workflows

Most AI nutrition tools rely on direct prompting.

That creates serious limitations:

• no domain grounding
• weak restriction handling
• no ingredient validation
• no compatibility logic
• no retrieval layer
• no fallback resilience
• no explainability

For healthcare-adjacent workflows, this is not enough.

NutriPrompt addresses this through layered AI architecture.

It is built to demonstrate how production-grade AI systems should be designed.

I built NutriPrompt to explore how AI systems can 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 think about AI product design.

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

NutriPrompt transforms structured user information into personalized nutrition workflows using:

• goal understanding
• digestive symptoms
• food preferences
• dietary restrictions
• budget logic
• lifestyle context

The AI output includes:

• weekly meal planning
• compatibility-aware recommendations
• practical tupper adaptation
• explainable profile tags
• downloadable PDF

NutriPrompt converts generated plans into:

• organized shopping lists
• category grouping
• optimized planning
• actionable execution

This transforms generation into utility.

NutriPrompt includes a dedicated public Streamlit demo to expose the internal AI workflow.

Designed for:

• technical interviews
• AI architecture walkthroughs
• product demos
• recruiter exploration
• portfolio storytelling

Structured nutrition intake.

Full workflow visibility.

Ingredient extraction and incompatibility detection.

Weekly nutrition planning.

Architecture and workflow observability.

Interactive explainability assistant:

• explains generated plans
• validates restrictions
• reviews ingredients
• provides contextual reasoning

This transforms NutriPrompt into an interactive AI decision-support layer.

Generate structured plans based on:

• goals
• restrictions
• symptoms
• preferences
• routines
• budget
• activity levels

NutriPrompt retrieves nutrition rules before generation.

Examples:

• Low FODMAP guidance
• gluten-free substitutions
• lactose-free alternatives
• digestive-safe patterns
• shopping optimization

This reduces hallucinations and improves consistency.

NutriPrompt analyzes:

• product labels
• pantry inventories
• ingredient lists
• nutrition PDFs
• fridge scans

OCR outputs are validated against user restrictions.

NutriPrompt evaluates:

• user restrictions
• retrieved rules
• OCR-detected ingredients

to detect conflicts before recommendation.

Generated plans are transformed into:

• categorized shopping lists
• planning-friendly structures
• execution-ready outputs

This reduces friction between recommendation and action.

NutriPrompt implements provider resilience:

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-ended prompting
• Retrieval before inference
• Validation before recommendation
• Explainability over black-box generation
• Provider abstraction for portability
• Separation between core product and public demo layer
• Structured outputs for downstream usability

These decisions reflect product-oriented AI architecture principles.

This project showcases:

• Prompt Engineering
• Retrieval-Augmented Generation (RAG)
• OCR Pipelines
• Multi-provider orchestration
• Explainable AI workflows
• Rule-based reasoning
• Validation layers
• Shopping intelligence
• Product-oriented AI architecture
• Service-oriented design
• Fallback resilience
• Structured AI outputs

NutriPrompt/
├── nutriprompt_app/
│   ├── services/
│   │   ├── ai/
│   │   ├── nutrition/
│   │   ├── profiles/
│   │   ├── rag/
│   │   ├── vision/
│   │   └── presentation/
│   ├── templates/
│   ├── tests/
│   └── views.py
│
├── streamlit_demo/
│   └── app.py
│
├── docs/
│   ├── screenshots/
│   └── streamlit_demo/

Current automated validation includes:

• prompt generation
• knowledge retrieval
• validation logic
• OCR processing
• context injection
• structured outputs
• shopping generation

Run:

python manage.py test

Current suite:

17 automated tests passing

git clone https://github.com/beatriangu/NutriPrompt.git
cd NutriPrompt

python3 -m venv venv
source venv/bin/activate

python -m pip install --upgrade pip
pip install -r requirements.txt

python manage.py runserver

Open:

http://127.0.0.1:8000/

streamlit run streamlit_demo/app.py

NutriPrompt provides informational guidance only.

It does not replace medical, nutritional or healthcare advice.

Outputs should be reviewed by qualified professionals where appropriate.

Bea Lamiquiz

🌐 Portfolio: https://bchill.net
💻 GitHub: https://github.com/beatriangu
💼 LinkedIn: https://www.linkedin.com/in/bealamiquiz/

If you’re working on AI products, LLM systems, RAG pipelines or GenAI workflows, I’d love to connect.

Open to:

• AI product conversations
• technical collaboration
• architecture discussions
• generative AI opportunities