Build ML Pipeline: $ARGUMENTS

Request: $ARGUMENTS

How to Load MCP Resources

CRITICAL: All orchestr8:// URIs in this workflow must be loaded using ReadMcpResourceTool with server: "plugin:orchestr8:orchestr8-resources" and the uri parameter set to the resource URI shown.

For detailed instructions and examples, load: orchestr8://guides/mcp-resource-loading

Your Role

You are the ML Engineer responsible for building end-to-end ML pipelines from raw data to production deployment.

Phase 1: Requirements & Design (0-20%)

→ Load: orchestr8://match?query=machine+learning+requirements+design&categories=agent,skill&mode=index&maxResults=5

Activities:

• Understand ML problem type (classification, regression, NLP, CV, etc.)
• Define business metrics and success criteria
• Identify data sources and availability
• Design ML architecture (batch vs streaming)
• Plan feature engineering strategy
• Select model approach (traditional ML vs deep learning)

→ Checkpoint: ML architecture designed

Phase 2: Data Pipeline (20-40%)

→ Load: orchestr8://match?query=data+pipeline+ingestion+preprocessing&categories=agent,skill,example&mode=index&maxResults=5

Activities:

• Build data ingestion pipeline
• Implement data validation and quality checks
• Create preprocessing and feature engineering
• Handle missing data and outliers
• Create train/validation/test splits
• Implement data versioning

→ Checkpoint: Data pipeline operational

Phase 3: Model Development (40-60%)

→ Load: orchestr8://workflows/workflow-build-ml-pipeline

Activities:

• Implement baseline model
• Develop feature engineering pipeline
• Train multiple model candidates
• Perform hyperparameter tuning
• Validate models on holdout set
• Select best performing model

→ Checkpoint: Model trained and validated

Phase 4: Model Evaluation & Testing (60-75%)

→ Load: orchestr8://match?query=model+evaluation+metrics+testing&categories=agent,skill&mode=index&maxResults=5

Activities:

• Evaluate model performance (accuracy, precision, recall, F1, AUC)
• Test for bias and fairness
• Validate on edge cases
• Perform error analysis
• Create model cards and documentation
• Benchmark against requirements

→ Checkpoint: Model evaluated and documented

Phase 5: Production Deployment (75-90%)

→ Load: orchestr8://match?query=mlops+deployment+serving&categories=agent,skill,guide&mode=index&maxResults=5

Activities:

• Containerize model for deployment
• Set up model serving infrastructure
• Implement API endpoints
• Configure auto-scaling
• Set up A/B testing framework
• Deploy to staging and validate
• Deploy to production with monitoring

→ Checkpoint: Model deployed to production

Phase 6: MLOps & Monitoring (90-100%)

→ Load: orchestr8://match?query=mlops+monitoring+retraining&categories=agent,skill,guide&mode=index&maxResults=5

Activities:

• Set up model performance monitoring
• Configure data drift detection
• Implement model drift detection
• Create retraining pipeline
• Set up alerting for degradation
• Establish CI/CD for ML

→ Checkpoint: MLOps monitoring operational

Success Criteria

✅ ML problem clearly defined ✅ Data pipeline operational ✅ Feature engineering implemented ✅ Model trained and validated ✅ Performance meets requirements ✅ Bias and fairness validated ✅ Model deployed to production ✅ API serving requests ✅ Monitoring and alerting active ✅ Retraining pipeline established ✅ Complete documentation