Capstone Project: Build a Functional Quantum AI Model

Table of Contents

1. Project Overview
2. Objective and Problem Definition
3. Tools and Environment Setup
4. Dataset Selection and Preprocessing
5. Feature Encoding Strategy
6. Quantum Circuit Design
7. Model Architecture (Hybrid Classical-Quantum)
8. Training and Optimization Loop
9. Evaluation Metrics and Validation Strategy
10. Running on Real Hardware (Optional)
11. Integration with API or UI
12. Logging and Experiment Tracking
13. Model Deployment Strategy
14. Security and Access Management
15. Benchmarking Against Classical Models
16. Scalability and Performance Tuning
17. Documentation and Codebase Structure
18. Final Presentation and Demo Plan
19. Submission Checklist
20. Conclusion

1. Project Overview

This capstone involves designing, training, evaluating, and deploying a fully functional Quantum AI model for a real-world task such as classification, recommendation, or regression using Qiskit or PennyLane.

2. Objective and Problem Definition

• Choose a problem domain: e.g., sentiment analysis, fraud detection
• Define success metrics: accuracy, AUC, latency
• State hypothesis: How will QML improve or complement classical ML?

3. Tools and Environment Setup

• PennyLane / Qiskit / TensorFlow Quantum
• Python 3.9+, Jupyter Notebook, VS Code
• Cloud backends: IBM Quantum, Amazon Braket (optional)

4. Dataset Selection and Preprocessing

• Use UCI, Kaggle, or custom dataset
• Normalize, reduce dimensions to 2–8 features
• Split into train/validation/test sets

5. Feature Encoding Strategy

• Angle Encoding
• Amplitude Encoding
• Basis Encoding
• Data re-uploading if needed

6. Quantum Circuit Design

• Feature Map: ZZFeatureMap, RealAmplitude
• Ansatz: TwoLocal, HardwareEfficientAnsatz
• Design for shallow circuits and noise resilience

7. Model Architecture (Hybrid Classical-Quantum)

• Classical frontend (dense or CNN)
• Quantum core for decision making
• Classical post-processing for final activation/output

8. Training and Optimization Loop

• Optimizers: SPSA, COBYLA, Adam, RMSProp
• Loss function: MSE, cross-entropy
• Use parameter-shift rule for gradients

9. Evaluation Metrics and Validation Strategy

• Accuracy, F1 score, precision, recall
• Use k-fold validation
• Compare with classical SVM/MLP

10. Running on Real Hardware (Optional)

• IBM Q Experience: use small backends like ibmq_quito
• Amazon Braket: IonQ or Rigetti
• Submit transpiled circuits with shot configuration

11. Integration with API or UI

• REST API with FastAPI or Flask
• Optional web frontend using Streamlit or React
• Expose /predict endpoint for inference

12. Logging and Experiment Tracking

• Use MLflow or Weights & Biases
• Track parameters, metrics, hardware used, and versions

13. Model Deployment Strategy

• Containerize with Docker
• Deploy on Heroku, AWS Lambda, or GCP Cloud Run
• Set up CI/CD using GitHub Actions

14. Security and Access Management

• Protect API with tokens or OAuth
• Use env files to store quantum backend credentials
• Rate-limit usage if public

15. Benchmarking Against Classical Models

• Train equivalent classical models (e.g., SVM, RandomForest)
• Compare performance and latency
• Discuss quantum speedups or tradeoffs

16. Scalability and Performance Tuning

• Minimize number of qubits
• Batch predictions
• Use cache for encoded inputs

17. Documentation and Codebase Structure

• README.md, requirements.txt, notebooks/, src/, api/, data/
• Include usage examples and training logs

18. Final Presentation and Demo Plan

• Present model architecture and rationale
• Show live or recorded demo of inference
• Visualize circuit and accuracy graphs

19. Submission Checklist

• [ ] Code repository (GitHub/Bitbucket)
• [ ] Final report (PDF or markdown)
• [ ] Trained model (simulated or QPU)
• [ ] Video demo or slide deck
• [ ] Deployment link (if any)

20. Conclusion

This capstone provides hands-on experience building and deploying a quantum AI model using modern hybrid techniques. By the end, students will demonstrate proficiency in QML pipeline design, circuit optimization, API deployment, and cross-model evaluation.