LiveMore: AI-Powered Health ROI Simulator (MVP)

Product Vision: An AI-powered platform that demonstrates the "Return on Investment" of healthy lifestyle choices through explainable biological age prediction.

Current Status: ✅ Model artifacts trained | 🏗️ Streamlit app development in progress
Timeline: 20-day sprint to thesis defense (Week 1 complete)

---

🎯 Project Overview

This repository represents a strategic pivot from building a complex production architecture to delivering a rapid-prototyping MVP that validates a core product hypothesis: Can Explainable AI (XAI) effectively communicate "Health ROI" to users?

The Pivot (October 28, 2025): After multiple scientific chaos injection attempts (Issues #49-50), we adopted a business-pragmatic approach focused on demonstrating value within thesis constraints. The result: a simplified dataset with explicit non-linear patterns where Random Forest meaningfully outperforms Linear Regression (+3-4% R² improvement).

The thesis defense will be presented as a "Startup Pitch" for "LiveMore," demonstrating the ability to:

• ✅ Generate defensible synthetic biological aging data (proprietary methods)
• ✅ Build ML models with explainable predictions (SHAP-based XAI)
• ✅ Validate product hypotheses quickly using modern prototyping (Streamlit)
• ✅ Balance scientific rigor with business practicality (MVP mindset)

🚀 Current Status

Sprint Timeline: 20 days to thesis defense (Day 6/20)
Current Phase: ✅ Week 1 Complete - Data & Models Ready | 🏗️ Week 2 Starting - Streamlit Development
Main Deliverable: antiaging-mvp/streamlit_app/ - Interactive Streamlit demo

Week 1 Achievements (Oct 22-28):

• ✅ Generated datasets_livemore_mvp/ with business-focused non-linear patterns
• ✅ Trained Random Forest model (R²=0.95, MAE=2.02 years on training)
• ✅ Created SHAP explainer for model interpretability
• ✅ Saved model artifacts to antiaging-mvp/streamlit_app/app_model/
• ✅ Documented pivot rationale and business approach

Strategic Focus

The project is currently focused on the Streamlit MVP. The production architecture (FastAPI backend, React frontend, ONNX export, Docker deployment) has been moved to /legacy and represents the post-validation roadmap.

Why this approach?

• ✅ Demonstrates technical leadership and product thinking
• ✅ Validates core hypothesis before heavy engineering investment
• ✅ Delivers working demo within thesis constraints
• ✅ Preserves future scalability path (see docs/FUTURE_ROADMAP_POST_TCC.md)

📂 Repository Structure

anti-aging-epigenetics-ml-app/
├── antiaging-mvp/
│   └── streamlit_app/           # 🎯 MAIN DELIVERABLE - Streamlit MVP
│       ├── app.py               # Main application (TO BE BUILT)
│       ├── app_model/           # ✅ Trained artifacts
│       │   ├── livemore_rf_v2.joblib      # Random Forest model
│       │   ├── livemore_scaler_v2.joblib  # Feature scaler
│       │   ├── livemore_explainer_v2.pkl  # SHAP explainer
│       │   └── model_metadata.json        # Model info
│       ├── requirements.txt     # Dependencies
│       └── README.md            # Setup instructions
│
├── ml_pipeline/                 # 🔬 ML Development Pipeline
│   ├── data_generation/         # Data generation engine
│   │   ├── generator_mvp_simple.py        # ✅ Business-focused generator
│   │   ├── generator_v2_biological.py     # Scientific chaos engine
│   │   ├── datasets_livemore_mvp/         # ✅ ACTIVE dataset
│   │   └── datasets_chaos_v2/             # Attempted scientific approach
│   ├── train_model_mvp.py       # ✅ Model training script
│   ├── quick_validation_mvp.py  # ✅ Rapid RF vs Linear validation
│   └── models/                  # Model evaluation utilities
│
├── notebooks/                   # Analysis & validation notebooks
│   └── baseline_figures/        # Visualization outputs
│
├── legacy/                      # 📦 ARCHIVED - Post-MVP production code
│   ├── backend_fastapi_archive/
│   ├── frontend_react_archive/
│   ├── datasets_chaos_v1_invalid/  # Failed attempt
│   └── notebooks_archive/
│
└── docs/                        # 📚 Comprehensive documentation
    ├── PIVOT.md                 # Strategic pivot rationale
    ├── PROJECT_STATUS_OCT_2025.md  # Detailed status & findings
    ├── INDEX.md                 # Documentation navigation
    └── FUTURE_ROADMAP_POST_TCC.md  # Post-thesis plans

🎓 The "LiveMore" Product Concept

Problem: People don't understand how lifestyle choices impact their biological aging

Solution: An AI-powered simulator that:

1. Predicts biological age from lifestyle/genetic factors
2. Explains each factor's impact using SHAP (Explainable AI)
3. Presents results as "Health ROI" (business-friendly language)

Technical Approach:

• Synthetic biological aging data (proprietary "Chaos Engine")
• Random Forest model for non-linear pattern detection
• SHAP for transparent, interpretable explanations
• Streamlit for rapid iteration and user testing

🔬 Core Intellectual Property: Synthetic Aging Data Generation

The project's main technical innovation is the synthetic data generation engine (ml_pipeline/data_generation/), which creates business-relevant aging datasets with:

Scientific Approach (generator_v2_biological.py):

• 10 aging-related SNPs with Hardy-Weinberg equilibrium
• 20 CpG methylation sites mimicking epigenetic clocks
• 5-phase chaos injection (heavy-tailed noise, interactions, age-variance)
• Result: Too complex for MVP timeline, archived as research foundation

Business Approach (generator_mvp_simple.py - ACTIVE):

• ✅ Simplified 9-feature model (age, gender, 6 lifestyle factors, genetic risk)
• ✅ Explicit non-linear patterns (thresholds, exponentials, U-curves)
• ✅ Strong interaction effects (smoking×stress, exercise×diet)
• ✅ Demonstrates clear RF advantage over Linear Regression
• ✅ Suitable for MVP demonstration and thesis defense

Current Dataset: datasets_livemore_mvp/ - 5000 training samples, 3 test sets

🏃 Quick Start (Streamlit MVP)

Navigate to the main deliverable:

cd antiaging-mvp/streamlit_app/

Follow the setup instructions in antiaging-mvp/streamlit_app/README.md

Prerequisites:

• Python 3.8+
• Virtual environment (recommended)
• Trained model artifacts (generated via Issue #50 workflow)

📊 Development Approach

Current Sprint (20 Days)

Week 1: Data Foundation ✅ COMPLETE (Oct 22-28)

• ✅ Archive production architecture to /legacy
• ✅ Attempted Issue #49-50 (scientific chaos injection)
• ✅ Pivoted to business-pragmatic approach
• ✅ Generated datasets_livemore_mvp/ with meaningful RF advantage
• ✅ Trained Random Forest model (R²=0.95, 9 features)
• ✅ Created SHAP explainer and model artifacts

Week 2: MVP Implementation 🏗️ IN PROGRESS (Oct 29 - Nov 4)

• Build Streamlit UI with sidebar inputs (age, lifestyle factors)
• Implement prediction display with biological age result
• Add SHAP waterfall/force plots for explanation
• Polish UI with business-friendly "Health ROI" language
• Test end-to-end user flow

Week 3: Thesis Preparation ⏳ UPCOMING (Nov 5-11)

• Create pitch deck using Streamlit demo as centerpiece
• Document pivot rationale and technical decisions
• Prepare defense materials with business narrative
• Practice demo and Q&A scenarios

Post-Thesis Roadmap

See docs/FUTURE_ROADMAP_POST_TCC.md for the complete production architecture plan:

• FastAPI REST API backend
• React/Next.js frontend
• ONNX model export for production inference
• PostgreSQL for user data
• Docker deployment
• Authentication and security

�� Documentation

Strategic Documents (Read These First!)

• docs/PIVOT.md - The complete strategic plan (20-day sprint)
• docs/PROJECT_STATUS_OCT_2025.md - Current status and Issue #50 justification

Technical Documentation

• antiaging-mvp/streamlit_app/README.md - MVP setup and architecture
• ml_pipeline/data_generation/README.md - Data generation engine docs

Reference Documentation

• docs/FUTURE_ROADMAP_POST_TCC.md - Post-MVP production roadmap
• README_PROFESSORS.md - Academic presentation for thesis committee

🎯 Success Metrics

MVP Success Criteria

• Working Streamlit app with local deployment
• Random Forest model outperforms Linear Regression by >5%
• SHAP explanations clearly show factor importance
• UI uses business-friendly "Health ROI" language
• Complete thesis pitch deck with live demo

Technical Validation

• datasets_chaos_v2/ passes non-linearity validation
• Model artifacts (<50MB) load quickly in Streamlit
• User can input lifestyle factors and see instant results
• SHAP waterfall/force plots render correctly

🔐 What's NOT in Scope (For This Thesis)

The following were deliberately moved to /legacy to maintain sprint focus:

• ❌ FastAPI backend implementation
• ❌ React frontend development
• ❌ ONNX model export
• ❌ Docker containerization
• ❌ Database integration (PostgreSQL)
• ❌ User authentication (JWT)
• ❌ Production deployment
• ❌ Load testing and optimization

Rationale: These represent the post-validation engineering roadmap. Building them now would consume the thesis timeline without validating the core hypothesis.

🤝 Academic Context

Institution: Undergraduate Thesis in Biological Sciences
Focus: Demonstrating technical leadership through product validation
Approach: Startup methodology applied to academic research
Timeline: 20-day sprint to defense

For the academic/professor-oriented view, see README_PROFESSORS.md.

📝 License

See LICENSE file for details.

---

Last Updated: October 27, 2025
Project Phase: Streamlit MVP Sprint (Week 1 - Data Foundation)
Next Milestone: Issue #50 completion and datasets_chaos_v2/ validation