PillTrack: MLOps Producer Hub

---

Overview

PillTrack Producer Hub is a streamlined MLOps production pipeline for medicine pack identification. The system manages the end-to-end lifecycle of dataset ingestion, AI-assisted auditing, feature extraction, artifact versioning, and synchronization with production environments via AWS S3.

This hub is designed for AI Operators and MLOps Engineers who need reliability, traceability, and rapid deployment for vision-based healthcare systems.

graph LR
    subgraph Data_Entry [Data Ingestion]
        UI["app.py (Streamlit UI)"]
        Raw["Raw Dataset"]
    end

    subgraph Core_Engine [src/ Processing Logic]
        YOLO["YOLOv8 (Segmentation)"]
        DINO["DINOv2 (Embeddings)"]
        Rotate["4-Dir Rotation Logic"]
    end

    subgraph Management [Registry & Audit]
        Audit["analytics.py (PCA/Audit)"]
        DB["db.manager.py (.pkl / .json)"]
    end

    subgraph Infrastructure [Production Sync]
        Cloud["cloud_manager.py (AWS SDK)"]
        S3[("☁️ AWS S3 Bucket")]
    end

    %% Process Flow
    UI -->|Upload| Raw
    Raw --> YOLO
    YOLO --> DINO
    DINO --> Rotate
    Rotate --> DB
    
    DB --> Audit
    Audit -->|Suggestions| UI
    
    UI -->|Trigger Sync| Cloud
    Cloud <-->|Push/Pull Artifacts| S3

    %% Styling for Clean Look
    style Data_Entry fill:#f5f5f5,stroke:#333
    style Core_Engine fill:#e3f2fd,stroke:#1565c0
    style Management fill:#f1f8e9,stroke:#33691e
    style Infrastructure fill:#fff3e0,stroke:#e65100

Loading

---

Architecture & Core Modules

The system follows a Modular "Src-Layout" Design to ensure scalability, testability, and clean separation of concerns.

Core Logic (src/)

• engine.py: The AI Brain.

  • YOLOv8 Segmentation for precise object localization.
  • DINOv2 for state-of-the-art feature extraction.
  • Generates 4-directional rotation-invariant vectors (0°, 90°, 180°, 270°).

• analytics.py: AI Auditor.

  • Performs PCA (Principal Component Analysis) for visualization.
  • Calculates Euclidean Distances to detect confusion risks and outliers.
  • Provides actionable suggestions (Low Data, High Spread).

• cloud_manager.py: Cloud Integration.

  • Handles secure artifact synchronization with AWS S3.
  • Includes robust error handling and connection status checks.

• db_manager.py: Data Registry.

  • Manages local vector databases (.pkl) and metadata generation (.json).

• utils.py: Helpers.

  • Centralized configuration loading and path management.

Controller

• app.py: Streamlit Dashboard.
  • Serves as the UI/Controller, orchestrating the interaction between the user and the backend modules.

---

Key Features

• Production Synchronization: One-click Push/Pull of models and vector DBs to AWS S3.
• AI-Powered Audit: Automatically detects dataset health issues (Imbalance, Confusion Risk, High Variance).
• Robust Testing: Fully unit-tested (pytest) covering Cloud logic, AI Engine, and Data flow.
• Automated Feature Extraction: Auto-crop and rotate images to build robust embeddings.
• Secure: Environment variables management via .env to protect credentials.

---

Project Structure

.
├── app.py                  # Main Streamlit UI Controller
├── config.yaml             # System Configuration
├── pytest.ini              # Testing Configuration
├── requirements.txt        # Dependencies
├── .env                    # Secrets (Not committed)
├── src/                    # Source Code (Core Logic)
│   ├── __init__.py
│   ├── analytics.py        # Math & Stats Logic
│   ├── cloud_manager.py    # AWS S3 Handler
│   ├── db_manager.py       # File & DB Handler
│   ├── engine.py           # YOLO + DINOv2 Engine
│   └── utils.py            # Config & Path Helpers
├── tests/                  # Unit Tests
│   ├── __init__.py
│   ├── test_cloud.py       # Cloud Mock Tests
│   └── test_engine.py      # AI Engine Mock Tests
├── database/               # Local Vector DB & Logs
└── models/                 # Pre-trained Model Weights (.pt)

Getting Started

1️⃣ Prerequisites Python 3.9 or higher

AWS Account with S3 Access

2️⃣ Installation

# Clone repository
git clone https://github.com/sitta07/PillTrack-Producer-Pipeline.git

# Install dependencies
pip install -r requirements.txt

3️⃣ Configuration Create a .env file in the root directory:

S3_BUCKET_NAME=your-production-bucket
AWS_ACCESS_KEY_ID=your_access_key
AWS_SECRET_ACCESS_KEY=your_secret_key
AWS_DEFAULT_REGION=your-default-region

4️⃣ Verification (Run Tests) Ensure everything is working correctly before running the UI:

# Run all unit tests
python -m pytest

5️⃣ Execution Start the MLOps Dashboard:

streamlit run app.py

👨‍💻 Author

Sitta Boonkaew
AI Engineer Intern @ AI SmartTech

---

📄 License

© 2025 AI SmartTech. All Rights Reserved.