Md Samshad Rahman

Md Samshad Rahman

He/His

MSc Applied CS grad Dalhousie University 🎓 | Python backend | LLM fine-tuning | AWS/GCP/Docker | Data-driven problem-solver

AI-Powered Mindful Eating App

This project was a journey in creating a truly personalized mindful eating experience from the ground up. My goal was to blend cutting-edge AI with user-friendly design to make a real difference in how people approach their meals.

  • The Tech Backbone: I architected and built the entire system using React Native (Expo) for a smooth mobile experience, Python (FastAPI) for a speedy and reliable backend, PostgreSQL for robust data management, and integrated Ollama/UnslothAI for powerful on-device/local LLM capabilities.
  • Smarter AI with Fine-Tuned LLMs:
    • I took two Meta Llama 3.2 (3B) models and customized them using LoRA adapters on Google Colab (T4 GPUs) – this was key to the app’s intelligence.
    • One model cleverly predicts a user’s Big Five personality traits based on their questionnaire responses.
    • The other model crafts context-aware mindful eating tips, which users found highly relevant (around 88% in pilot tests!).
  • Crafting a Unique Knowledge Base: I meticulously curated a dataset of 1,500 mindful eating tips inspired by experts. These were carefully mapped across 10 eating behaviors and 5 personality types, then further refined with ChatGPT-4o, and validated by a registered dietitian and a psychologist to ensure quality and accuracy.
  • Powerful & Efficient Backend: I engineered a robust FastAPI backend with REST APIs that handle everything smoothly:
    • Secure user sign-up, login, and profile management (including demographics, dietary preferences, and personality insights).
    • Easy goal setting and food logging (users can even snap a photo or just type).
    • Instant, intelligent responses from the fine-tuned AI.
  • Full-Stack Design & Development: I designed and developed the mobile app, ensuring an intuitive flow for users to input their information, receive personalized AI feedback, and even chat with the AI for support.
  • Understanding Users Better: To deepen personalization, I also trained a Big Five personality classifier using the Symanto NLP API on a dataset of 2,467 essays. This helps the app tailor advice even more effectively.

What This Project Showcases:

  • End-to-End Development: From concept to a fully functional application.
  • Advanced AI/ML Skills: Fine-tuning large language models (Llama 3, LoRA), dataset curation, and practical LLM integration.
  • Full-Stack Proficiency: Expertise across the front-end (React Native), backend (FastAPI, Python), database (PostgreSQL), and MLOps (Ollama, UnslothAI).
  • User-Centric Design: Focusing on creating a helpful, personalized, and engaging experience.
  • Problem-Solving & Innovation: Combining diverse technologies to build a novel solution for promoting healthier habits.

Technologies I Mastered & Used: Python, FastAPI, PostgreSQL, React Native, Expo, Ollama, UnslothAI, LoRA, LLM Fine-Tuning, Google Colab, Symanto NLP API, Docker, AWS EC2.

Code on GitHub

System Architecture:

System Architecture
Figure: A look under the hood – the system architecture of the AI-Powered Mindful Eating Companion.