Advanced Projects

Complex, production-style portfolio projects. Completing a brief is not the same as SRE-hardened production (auth, multi-AZ, SLOs, on-call). These nine projects ship as detailed README briefs and capstone-style specs. You design and implement the full pipeline.

Time series: Project 3 overlaps with Module 15 and intermediate project 6, pick one path: TIME_SERIES_LEARNING_PATH.md.

Prerequisites

Before starting these projects, you should have completed:

• All Intermediate Projects
• Module 09: Neural Networks Basics
• Module 10: Deep Learning Frameworks
• Module 11: Computer Vision (for CV projects)
• Module 12: Natural Language Processing (for NLP projects)
• Module 13: Model Deployment
• Module 14: MLOps Basics

CNN and RNN curriculum map (projects)

Projects

Project 1: Image Classification (CIFAR-10)

Difficulty: Advanced
Time: 1-2 weeks
Skills: CNNs, Transfer Learning, Data Augmentation, Model Optimization

Build a CNN to classify images into 10 categories with high accuracy.

What you'll learn:

• Advanced CNN architectures
• Transfer learning with pre-trained models
• Data augmentation strategies
• Model optimization techniques
• Handling complex image data

Dataset: CIFAR-10 (built into Keras)

Extensions:

• Achieve >90% accuracy
• Try different architectures (ResNet, EfficientNet)
• Deploy as web service
• Build mobile app integration

Project 2: Sentiment Analysis on Reviews

Difficulty: Advanced
Time: 1-2 weeks
Skills: NLP, RNNs/LSTMs, Transformers, Text Classification

Build a sentiment analysis system for product/movie reviews.

What you'll learn:

• Advanced text preprocessing
• Word embeddings (Word2Vec, GloVe)
• RNN/LSTM architectures
• Transformer models (BERT)
• Handling different text lengths

Dataset: IMDB Reviews or Amazon Reviews

Extensions:

• Multi-class sentiment (positive/neutral/negative)
• Aspect-based sentiment analysis
• Real-time API
• Model explainability

Project 3: Time Series Forecasting (Advanced)

Difficulty: Advanced
Time: 1-2 weeks
Skills: Time Series, LSTM, ARIMA, Prophet, Feature Engineering

Build advanced forecasting models for complex time series data.

What you'll learn:

• Advanced time series techniques
• LSTM for sequences
• Prophet for trend/seasonality
• Feature engineering for time series
• Multi-step ahead forecasting

Dataset: Stock prices, energy consumption, sales data

Extensions:

• Multiple time series
• External features
• Uncertainty quantification
• Real-time forecasting system

Project 4: LLM Chatbot & RAG System

Difficulty: Advanced
Time: 2-3 weeks
Skills: Modern LLMs, RAG, Vector Databases, LangChain, Prompt Engineering

Build a production-style RAG (Retrieval-Augmented Generation) system that ingests documents, stores embeddings in vector databases, and answers questions using GPT-4/Llama 3 or other LLMs. Treat eval, citations, and cost logs as part of "done."

What you'll learn:

• Prompt engineering techniques
• Vector database integration (Pinecone, ChromaDB, or Weaviate)
• Document ingestion and chunking
• Embedding generation and storage
• RAG pipeline implementation
• LangChain/LangGraph for orchestration
• Evaluation of RAG systems
• Production deployment

Core Components:

1. Document Processing: PDF/document parsing and chunking
2. Embedding Generation: Convert text to vectors using embedding models
3. Vector Database: Store and retrieve relevant document chunks
4. LLM Integration: Use GPT-4, Llama 3, or open-source LLMs
5. RAG Pipeline: Combine retrieval with generation
6. Evaluation: Measure answer quality and relevance

Dataset:

• Custom PDFs/documents (technical docs, research papers, company knowledge base)
• Or use public datasets like Natural Questions

Tech Stack:

• LLM: OpenAI GPT-4, Anthropic Claude, or Llama 3 (via Hugging Face)
• Vector DB: Pinecone, ChromaDB, Weaviate, or FAISS
• Framework: LangChain or LangGraph
• Embeddings: OpenAI embeddings, sentence-transformers, or BGE models

Extensions:

• Multi-document RAG
• Citation and source tracking
• Streaming responses
• Chat history and context management
• Deploy to Hugging Face Spaces or AWS
• Add evaluation metrics (BLEU, ROUGE, semantic similarity)
• Implement query rewriting and query expansion

Project 5: Object Detection

Difficulty: Advanced
Time: 2-3 weeks
Skills: Computer Vision, Object Detection, YOLO, R-CNN, Transfer Learning

Detect and localize multiple objects in images.

What you'll learn:

• Object detection architectures
• YOLO (You Only Look Once)
• R-CNN family
• Bounding box regression
• Evaluation metrics (mAP)

Dataset: COCO Dataset or custom dataset

Extensions:

• Real-time detection
• Multiple object classes
• Video object detection
• Deploy for edge devices

Project 6: End-to-End ML Pipeline

Difficulty: Advanced
Time: 2-3 weeks
Skills: MLOps, CI/CD, Model Deployment, Monitoring, Full Stack

Build a complete, production-style ML system from data to a local or staging deploy. A README checklist is not a multi-region production launch.

What you'll learn:

• End-to-end ML pipeline
• Data versioning (DVC)
• Experiment tracking (MLflow)
• Model deployment (Docker, Kubernetes)
• Monitoring and logging
• CI/CD for ML

Project Components:

• Data pipeline
• Training pipeline
• Model registry
• API service
• Monitoring dashboard
• Automated retraining

Extensions:

• A/B testing framework
• Feature store
• Model explainability dashboard
• AutoML integration

Project 7: Generative Model (GAN/VAE)

Difficulty: Advanced
Time: 2-3 weeks
Skills: Generative Models, GANs, VAEs, Deep Learning

Build a generative model to create new images or text.

What you'll learn:

• Generative Adversarial Networks (GANs)
• Variational Autoencoders (VAEs)
• Training challenges
• Evaluation of generative models
• Applications

Dataset: CelebA, MNIST, or custom dataset

Extensions:

• Style transfer
• Image-to-image translation
• Text generation
• Conditional generation

Project 8: Model Explainability & Interpretability

Difficulty: Advanced
Time: 1-2 weeks
Skills: SHAP, LIME, Model Interpretation, Explainable AI

Build an explainable ML workflow using SHAP, LIME, and related interpretability tools.

What you'll learn:

• SHAP values (Tree, Kernel, Deep)
• LIME for local explanations
• Partial Dependence Plots
• Feature importance analysis
• Building explainable ML systems
• Regulatory compliance considerations

Dataset: Credit scoring, medical diagnosis, or loan approval datasets

Extensions:

• Text/image explanations
• Counterfactual explanations
• Fairness analysis
• Explanation quality metrics

Project 9: Model Deployment & Serving

Difficulty: Advanced
Time: 1-2 weeks
Skills: FastAPI, Docker, REST APIs, Cloud Deployment, Model Monitoring

Deploy a machine learning model as a production-shaped API demo with Docker, basic monitoring hooks, and clear gaps called out (auth, SLOs, on-call).

What you'll learn:

• Building production-shaped REST APIs with FastAPI (demo quality, not a blank check for production)
• Containerizing ML models with Docker
• Deploying models to cloud platforms
• Implementing model versioning
• Setting up monitoring and logging
• Handling errors and edge cases
• Optimizing for production

Dataset: Any trained model from previous projects

Extensions:

• A/B testing
• Model monitoring
• Feature store integration
• Kubernetes deployment

Capstone Blueprints (Optional, Portfolio-Grade)

These are industry-style capstone blueprints that are designed to be safe for a public repo (no PII, no committed datasets).

• Capstones Index →
• ML Engineer Capstone →
• LLM/RAG Capstone →
• Data/Analytics (SQL → ML) Capstone →

Project Structure

Each project should include:

project-name/
 README.md              # Comprehensive project documentation
 data/                  # Dataset and data processing scripts
 notebooks/             # Research and experimentation
 src/                   # Production code
    data/
    models/
    training/
    evaluation/
    deployment/
 tests/                 # Comprehensive test suite
 config/                # Configuration files
 models/                # Saved models and checkpoints
 results/               # Results, papers, presentations
 docs/                  # Documentation
 docker/                # Dockerfiles
 .github/               # CI/CD workflows
 requirements.txt       # Dependencies
 setup.py               # Package setup
 LICENSE                # License file

Tips for Success

1. Research First: Read papers and understand current strong baselines
2. Start Simple: Build baseline, then add complexity
3. Version Control: Use Git, DVC, MLflow extensively
4. Document Everything: Code, experiments, decisions
5. Test Thoroughly: Unit tests, integration tests
6. Deploy: Actually deploy to production (even if simple)
7. Monitor: Set up monitoring and logging
8. Present: Write a clear presentation or report

Learning Outcomes

After completing these projects, you should be able to:

• Build production-style ML systems you can demo and extend
• Apply strong current techniques with honest eval
• Handle complex, real-world problems
• Deploy and monitor ML models
• Work with large datasets
• Optimize models for production
• Explain and present ML solutions

Additional Resources

• Papers with Code: Latest research
• arXiv: Research papers
• GitHub Awesome Lists: Curated resources
• Kaggle Competitions: Advanced challenges

Portfolio Building

These projects help you:

• Build portfolio evidence
• Show skills to employers (no hire guarantee)
• Practice open-source style READMEs
• Draft blog posts or meetup talks

Ready? Pick one project that matches your target role and start building. Keep the README honest about what you measured.