Buildable ML/DL Fellowship 🤖🚀

Welcome to the Buildable ML/DL Fellowship repository!
This is a 3-month immersive journey into Machine Learning (ML) and Deep Learning (DL), designed to build strong foundations in AI while practicing hands-on coding and real-world projects.

This repository showcases my complete learning progression—from Python fundamentals to building production-ready agentic AI systems—through weekly assignments, comprehensive projects, and a capstone project.

---

🌍 About Buildables

Buildables is a learning community in Pakistan focused on empowering students and developers in AI, ML, and Deep Learning.
Through collaborative projects, mentorship, and peer discussions, it encourages learners to explore intelligent systems and apply their skills to real-world problems.

---

📚 What I Learned

Throughout this fellowship, I gained practical skills and theoretical foundations in ML/DL, progressing from basic programming to advanced AI systems. Here's my learning journey organized from foundational concepts to advanced implementations:

🐍 Python Fundamentals & Programming Logic

• Writing clean, efficient Python code with proper structure
• Data structures: Mutable vs Immutable (lists, tuples, dictionaries)
• Input validation, error handling, and conditional logic
• Function creation, lambda functions, and functional programming
• Statistical operations (mean, median, max, min)
• File handling and data manipulation

📊 Data Handling & Preprocessing

• Working with NumPy and Pandas for data manipulation
• Handling missing values with multiple imputation strategies
• Data cleaning, transformation, and normalization
• Feature encoding (one-hot encoding, label encoding)
• Feature scaling and standardization
• Data augmentation techniques with Gaussian noise
• Creating reproducible data pipelines

📈 Exploratory Data Analysis (EDA)

• Statistical summarization and descriptive analysis
• Data visualization with Matplotlib and Seaborn
• Creating histograms, scatter plots, box plots, and correlation heatmaps
• Identifying patterns, outliers, and anomalies
• Domain-specific data exploration (weather, automotive, real estate, agriculture)

🔧 Feature Engineering

• Creating derived features from existing data
• Text feature extraction and processing
• Geospatial feature engineering
• Feature interaction and polynomial features
• Feature selection and dimensionality reduction
• Domain knowledge application for better features

🤖 Machine Learning Foundations

• Supervised Learning: Regression and Classification
• Model Training: Train-test splitting, stratified sampling
• Algorithms Implemented:
  • Linear/Logistic Regression
  • Decision Trees
  • Random Forest (ensemble learning)
  • Support Vector Machines
  • K-Nearest Neighbors
• Model Evaluation: Accuracy, Precision, Recall, F1-Score, ROC-AUC, Confusion Matrix
• Cross-Validation: K-fold validation for robust performance estimation
• Model Serialization: Saving models with joblib for deployment

🧠 Deep Learning Essentials

• Neural network architecture design
• Artificial Neural Networks (ANNs): Fully connected layers, activation functions
• Convolutional Neural Networks (CNNs): Conv2D, MaxPooling, feature extraction
• Understanding backpropagation and gradient descent
• Optimizer selection (Adam, SGD)
• Loss functions (Binary Crossentropy, Categorical Crossentropy)
• Regularization techniques (Dropout, Early Stopping)
• Model callbacks (EarlyStopping, ModelCheckpoint)
• Training loops and epoch management

🖼️ Computer Vision & Image Classification

• Image preprocessing and normalization
• Building CNNs from scratch for image classification
• Transfer Learning: Fine-tuning pretrained models (ResNet50)
• Data augmentation (rotation, flipping, color jitter)
• Handling different image datasets:
  • MNIST (28x28 grayscale, 10 classes) → 98.64% accuracy
  • Cat vs Dog (150x150 RGB, binary) → 69% accuracy
  • Plant Disease (224x224 RGB, 15 classes) → 98.97% accuracy
• Performance comparison: ANN vs CNN architectures
• Training on GPU with Google Colab

🔍 Advanced ML Techniques

• End-to-end ML pipeline development
• Hyperparameter tuning strategies
• Handling class imbalance
• Ensemble methods and model stacking
• Feature importance analysis
• Model interpretability
• Binary and multi-class classification
• Regression modeling for continuous predictions

🤝 Collaboration & Version Control

• Git/GitHub workflows (branches, commits, pull requests)
• Repository organization and documentation
• Professional README creation
• Code commenting and documentation practices
• Collaborative development workflows

🤖 Agentic AI Systems & RAG Implementation

• Retrieval-Augmented Generation (RAG):
  • Vector embeddings with sentence-transformers
  • FAISS vector database for semantic search
  • Context retrieval and relevance ranking
  • LLM integration (Google Gemini API)
• Intent Classification: Query understanding and routing
• Multi-Tool Agent Systems: Intelligent tool selection
• Production AI Architecture: Modular, maintainable system design
• Multi-Modal AI: Combining text, images, and structured data

---

🎯 Key Projects & Achievements

📁 Synthetic Data Pipeline

Built an end-to-end ML pipeline with data generation, cleaning, augmentation, and model training.

• Models: Logistic Regression vs Random Forest
• Best Result: 84% accuracy with Random Forest (ROC-AUC: 0.89)
• Key Skills: Data augmentation, pipeline automation, artifact management

🌦️ Weather Classification System

Multi-class classification predicting weather types from meteorological features.

• Dataset: 11 features (temperature, humidity, wind speed, precipitation, etc.)
• Target: 4 weather types (Rainy, Sunny, Cloudy, Snowy)
• Models: Logistic Regression, Decision Tree, Random Forest
• Key Skills: Multi-class classification, categorical encoding, model comparison

🚗 Car Price Analysis

Regression analysis on Pakistani used car market data (PakWheels dataset).

• Features: Make, Model, Year, Engine CC, Mileage, Transmission
• Focus: Comprehensive EDA and preprocessing for price prediction
• Key Skills: Real-world data cleaning, regression preprocessing

🐱🐶 Image Classification: ANN vs CNN

Comparative study demonstrating CNN superiority for image tasks.

• MNIST Digits: ANN (97.74%) vs CNN (98.64%)
• Cat vs Dog: ANN (57.30%) vs CNN (69%)
• Key Finding: CNNs excel at spatial feature extraction
• Key Skills: Deep learning, model comparison, transfer learning basics

🏠 Airbnb Listing Analysis

Complete ML/DL project analyzing Albany, NY Airbnb data.

• Tasks: Price prediction (regression) + Popularity classification
• Features: Text extraction, geospatial engineering, categorical encoding
• Models: Traditional ML + Neural Networks
• Key Skills: Feature engineering, hybrid ML/DL approach, end-to-end project

🌾 Smart Farming Advisor (Final Capstone)

Production-ready agentic AI system for precision agriculture with three integrated tools.

1. Crop Recommendation Tool

• Algorithm: Random Forest with feature engineering
• Accuracy: 99.39% on 22 crop types
• Features: Soil nutrients (N, P, K), climate (temp, humidity, pH, rainfall)

2. Disease Detection Tool

• Architecture: ResNet50 (Transfer Learning)
• Accuracy: 98.97% on 15 plant diseases
• Training: Google Colab T4 GPU (33 minutes)
• Dataset: 20,639 images with data augmentation

3. RAG Q&A System

• Components: FAISS vector DB + Gemini LLM
• Hit Rate: 100% (perfect retrieval)
• MRR: 1.0 (best document always ranked first)
• Knowledge Base: 10 farming FAQ documents

Intelligent Agent

• Routing Accuracy: 100% (4/4 test queries)
• Functionality: Automatic intent classification and tool selection

Key Skills: Multi-modal AI, transfer learning, RAG, vector databases, agentic systems, production architecture

---

📊 Performance Highlights

Project	Model/Method	Metric	Achievement
Synthetic Pipeline	Random Forest	ROC-AUC	89. 89%
MNIST Classification	CNN	Accuracy	98.64%
Cat vs Dog	CNN	Accuracy	69%
Crop Recommendation	Random Forest	Accuracy	99.39% ✅
Disease Detection	ResNet50	Accuracy	98.97% ✅
RAG Q&A	FAISS + Gemini	Hit Rate@3	100% ✅
Agent Routing	Intent Classifier	Routing	100% ✅

---

🛠️ Technologies & Tools Mastered

Programming & Core Libraries

• Python 3. 8+, Jupyter Notebook, Google Colab

Data Science Stack

• NumPy, Pandas, Matplotlib, Seaborn

Machine Learning

• scikit-learn (Classification, Regression, Ensemble Methods)
• Feature Engineering, Cross-Validation, Model Evaluation

Deep Learning Frameworks

• TensorFlow/Keras (CNNs, ANNs, Callbacks)
• PyTorch (Transfer Learning, ResNet50)

Advanced AI

• FAISS (Vector Database)
• sentence-transformers (Embeddings)
• Google Gemini API (LLM Integration)

Development Tools

• Git/GitHub (Version Control, Branching, PRs)
• joblib (Model Serialization)
• Environment Management

---

📁 Repository Structure

Buildable-ML-DL-Fellowship/
├── Week_1/                          # Python fundamentals & logic
├── Week_3/                          # Data pipeline & ML basics
│   └── synthetic_data_pipeline_project/
├── Week_4/                          # Classification & Regression
│   ├── classification/              # Weather prediction
│   └── regression/                  # Car price analysis
├── Week_6/                          # Deep Learning (ANN vs CNN)
├── Week_7/                          # Airbnb analysis project
└── final-project-smart-farming-advisor/  # Capstone agentic AI system
    ├── src/
    │   ├── data/                    # Data preprocessing
    │   ├── models/                  # Model training
    │   ├── rag/                     # RAG pipeline
    │   ├── tools/                   # Three AI tools
    │   ├── agent/                   # Intelligent agent
    │   └── evaluation/              # Metrics & testing
    ├── models/                      # Trained models
    ├── outputs/                     # Results & visualizations
    └── notebooks/                   # Colab training notebooks

---

🌟 Fellowship Experience

This fellowship was uniquely flexible, providing:

• Self-paced learning with curated materials
• Hands-on weekly assignments focusing on practical skills
• 3 weeks for the final project allowing deep exploration
• Real-world datasets from Kaggle and public sources
• Mentorship and community support through Buildables

This repo documents my journey—it's more than just code, it's about thinking like a researcher, building like an engineer, and collaborating like a team player.

---

🚀 Key Learnings & Takeaways

✅ Progression from basics to advanced: Python → ML → DL → Production AI
✅ Hands-on implementation: 20+ models trained across various domains
✅ Real-world problem solving: Weather, automotive, real estate, agriculture
✅ Production-ready skills: 99%+ accuracy systems, modular architecture
✅ Multi-modal AI: Text, images, structured data integration
✅ Best practices: Version control, documentation, testing, evaluation

---

🎓 What's Next?

• Advanced NLP and transformer architectures
• Model deployment with FastAPI and Docker
• MLOps practices and experiment tracking
• Reinforcement Learning fundamentals
• Contributing to open-source AI projects

---

🤝 Connect

Let's continue the Buildables spirit:

• Share ideas 💡
• Collaborate on projects 🤝
• Grow together as an AI community 🌍

GitHub: @juni2003
Email: juni.xatti@gmail.com

Happy Learning & Coding! 🚀