Skip to Content
18 Low-Code AI Tools07 Assignment

Assignment: Low-Code ML Application

Phase 18: Low-Code AI Tools
Scope: End-to-end app build with optional extra polish
Suggested Pace: 1-2 weeks, depending on deployment and polish

📋 Assignment Overview

Build and deploy a complete machine learning application using low-code tools. Your application should include AutoML model training, an interactive interface (Gradio or Streamlit), and deployment to Hugging Face Spaces.

🎯 Learning Objectives

  • Apply AutoML for model development
  • Build user-friendly ML interfaces
  • Deploy applications to the cloud
  • Integrate multiple low-code tools
  • Follow best practices for production ML

📝 Requirements

Part 1: Dataset and Problem

Choose ONE of the following:

Option A: Classification Problem

  • Predict customer behavior (churn, purchase, click)
  • Medical diagnosis
  • Sentiment analysis
  • Fraud detection

Option B: Regression Problem

  • Price prediction (house, stock, product)
  • Demand forecasting
  • Performance estimation
  • Time series prediction

Deliverables:

  • Dataset description (source, size, features)
  • Problem statement
  • Success metrics
  • Exploratory data analysis (EDA)

Self-check:

  • Clear problem definition
  • Appropriate dataset
  • EDA with visualizations

Part 2: Model Training with AutoML

Use PyCaret, FLAML, or H2O AutoML to train your model.

Requirements:

  • Data preprocessing (handle missing values, encode categoricals)
  • Train multiple models using AutoML
  • Compare at least 5 different algorithms
  • Tune hyperparameters of best model
  • Evaluate performance (accuracy, precision, recall, etc.)
  • Save final model

Deliverables:

  • Jupyter notebook with AutoML pipeline
  • Model comparison results
  • Performance metrics on test set
  • Saved model file (.pkl or .h5)

Self-check:

  • Proper data preprocessing
  • Model comparison and selection
  • Performance evaluation
  • Model saved correctly

Part 3: Interactive Interface

Build EITHER a Gradio interface OR a Streamlit app.

Option A: Gradio Interface

Requirements:

  • Clear input components for all features
  • Prediction output with confidence score
  • At least 3 example inputs
  • Custom theme and styling
  • Error handling for invalid inputs

Option B: Streamlit App

Requirements:

  • Sidebar for input controls
  • Main area for predictions and visualizations
  • Session state for user interactions
  • Caching for model loading
  • Interactive charts (plotly or matplotlib)

Both Options Must Include:

  • Clear instructions for users
  • Feature descriptions/tooltips
  • Actionable output (not just raw predictions)
  • Professional appearance

Deliverables:

  • app.py file with complete interface
  • requirements.txt with all dependencies
  • Screenshots of working interface

Self-check:

  • User-friendly design
  • Proper functionality
  • Error handling and edge cases
  • Professional appearance

Part 4: Deployment

Deploy your application to Hugging Face Spaces.

Requirements:

  • Create Hugging Face account
  • Set up new Space (Gradio or Streamlit SDK)
  • Upload all necessary files:
    • app.py
    • requirements.txt
    • README.md (with frontmatter)
    • Trained model
  • Ensure app runs successfully
  • Test all features in deployed version

Deliverables:

  • Public Hugging Face Space URL
  • README.md with:
    • App description
    • Usage instructions
    • Model information
    • Example inputs
  • Working deployed application

Self-check:

  • Successful deployment
  • Complete README
  • App functionality
  • Professional presentation

Part 5: Documentation and Report

Create a comprehensive report including:

  1. Introduction (1-2 pages)

    • Problem description
    • Dataset overview
    • Approach summary

  2. Methods (2-3 pages)

    • AutoML platform choice and justification
    • Model comparison process
    • Final model selection reasoning
    • Interface design decisions

  3. Results (1-2 pages)

    • Model performance metrics
    • Comparison table
    • Visualizations of results
    • Discussion of strengths/limitations

  4. Deployment (1 page)

    • Deployment process
    • Challenges faced
    • Solutions implemented

  5. Conclusion (1 page)

    • Summary of achievements
    • Lessons learned
    • Future improvements

Deliverables:

  • PDF report (5-10 pages)
  • Well-organized sections
  • Visualizations and tables
  • References (if applicable)

Self-check:

  • Clarity and organization
  • Technical accuracy
  • Completeness

🌟 Optional Extensions

Optional Stretch Ideas:

  1. Advanced Features

    • SHAP explanations for predictions
    • Batch prediction with file upload
    • Model comparison dashboard
    • Real-time data visualization
    • A/B testing different models

  2. Additional Deployment

    • Deploy to multiple platforms (Docker, Cloud Run, etc.)
    • Set up CI/CD pipeline
    • Add monitoring/logging

  3. Outstanding Quality

    • Exceptional documentation
    • Production-ready code quality
    • Comprehensive error handling
    • Accessibility features

📤 Deliverables & Packaging

Prepare the following:

  1. GitHub Repository containing:

    • Jupyter notebook(s) for model training
    • app.py for interface
    • requirements.txt
    • README.md
    • Saved model file
    • Any additional code files

  2. Hugging Face Space URL

    • Working deployed application
    • Complete README on Space

  3. Report (PDF)

    • Following structure outlined above

  4. Project Summary with:

    • Project name
    • GitHub repository URL
    • Hugging Face Space URL
    • Report PDF
    • Brief summary (200 words)

Suggested Packaging:

  • Keep all materials in one well-organized repository or archive
  • Use a clear project name such as phase18_low_code_assignment/
  • Make it easy for another learner or reviewer to run the app and inspect the results

✅ Self-Review Guide

ComponentPointsCriteria
Dataset & Problem15Clear problem, appropriate data, EDA
AutoML Training25Proper workflow, model comparison, evaluation
Interface30User-friendly, functional, professional
Deployment20Working app, complete README, accessible
Documentation10Clear, organized, comprehensive report
Bonus+15Advanced features, extra deployment, quality
TOTAL100 (+15)

💡 Tips for Success

Technical Tips:

  1. Start simple - Get basic version working first
  2. Test locally - Debug before deploying
  3. Version control - Commit frequently
  4. Pin dependencies - Specify exact versions
  5. Handle errors - Add try-except blocks

Design Tips:

  1. User-first - Think about user experience
  2. Clear labels - Make inputs obvious
  3. Provide examples - Help users understand
  4. Show confidence - Display prediction uncertainty
  5. Add context - Explain what predictions mean

Deployment Tips:

  1. Check requirements - Test all dependencies
  2. Optimize model - Reduce size if possible
  3. Test thoroughly - Try edge cases
  4. Monitor logs - Watch build process
  5. Update README - Keep documentation current

📚 Resources

AutoML Platforms:

Interface Tools:

Deployment:

Example Projects:

  • Browse Hugging Face Spaces for inspiration
  • Check course notebooks for patterns
  • Review PyCaret examples

❓ FAQ

Q: Can I use my own dataset?
A: Yes! As long as it meets the requirements (appropriate size, clear problem).

Q: Which AutoML platform should I use?
A: PyCaret is recommended for beginners. FLAML for faster training. H2O for larger datasets.

Q: Gradio or Streamlit?
A: Gradio is simpler for basic interfaces. Streamlit offers more customization.

Q: What if my model is too large to upload?
A: Use model compression, reduce features, or use Git LFS for large files.

Q: How long should my app take to load?
A: Aim for < 30 seconds initial load, < 2 seconds per prediction.

Q: Can I work in a group?
A: Individual work is the cleanest path. If you collaborate, keep ownership of the model, interface, and deployment decisions explicit.

Q: What if deployment fails?
A: Check logs, verify requirements.txt, test locally first, ask for help.

🎯 Example Project Ideas

Beginner Level:

  • Iris flower classifier
  • Boston house price predictor
  • Titanic survival predictor
  • Wine quality estimator

Intermediate Level:

  • Customer churn predictor
  • Loan default predictor
  • Product recommendation system
  • Sentiment analyzer

Advanced Level:

  • Real-time fraud detection
  • Medical diagnosis assistant
  • Stock price forecaster
  • Image classification system

🏆 Suggested Build Timeline

  • Stage 1: Dataset selection and EDA
  • Stage 2: Model training and evaluation
  • Stage 3: Interface development
  • Stage 4: Deployment and documentation

Use the stages above as a pacing guide if you are studying independently.

📧 Getting Help

  • Discussion Forum: GitHub Discussions 
  • Technical Issues: Include logs, screenshots, deployment errors, and your requirements.txt.
  • Best workflow: Test locally before pushing to Hugging Face Spaces.

Good luck! Build something amazing! 🚀

Last updated on
06 End To End Project08 Challenges