Artificial intelligence project management

Your job is to investigate one failed or problematic AI project and analyze what went wrong. Choose from a viewport example (such as Zillow, Tay, Amazon Haining tool) or find another case that is publicly known.

The report should analyse:

• Project objectives and initial context
• Errors (technical, organisational, ethical, etc.)
• Which stages of life cycle or feasibility assessment were weak or omitted
• What could have been done differently in terms of project management or risk analysis
• What lessons can be learned from future AI projects in this example

For any of these topics, try to find real world examples / projects / blogposts

Option 1: Data Strategy and Quality Impact

"From Garbage In to Gold Out: How Data Strategy Decisions Shape AI Project Success"

Students should examine the interconnected relationship between data quality, team dynamics, and experimental outcomes. They should address:

• How poor data strategy decisions cascade through team workflows
• The role of different team members in maintaining data quality
• Trade-offs between synthetic data generation and real data collection
• How data versioning and metadata management affect experiment reproducibility

Option 2: The Build vs. Buy Decision

"Reinventing the Wheel or Rolling Your Own: Strategic Decisions in AI Tool Selection"

Students analyze the trade-offs between building custom solutions versus using existing tools across the AI project lifecycle:

• Custom labeling tools vs. commercial platforms (Labelbox, Amazon SageMaker)
• In-house experiment tracking vs. MLflow/Weights & Biases/ClearML
• Self-hosted infrastructure vs. cloud solutions
• Building internal data pipelines vs. using managed services
• Impact of these decisions on team productivity, skills requirements, and project outcomes

Choose ONE of the following topics or propose your own related topic for approval:

Pipeline Architecture & Design

• "Building Robust ML Pipelines: A Critical Analysis" - Analyze the key components of ML pipelines, discussing trade-offs between modularity, scalability, and complexity. Include real-world examples of pipeline failures and how robust design could prevent them.
• "Orchestration Tools Comparison: When Simple Scripts Beat Complex Systems" - Compare Apache Airflow, Dagster, and simple Python scripts for different use cases. Discuss when each approach is most appropriate and the hidden costs of over-engineering.

Production Optimization & Scaling

• "Model Optimization for Production: Beyond Accuracy Metrics" - Examine techniques like quantization, pruning, and efficient architectures. Analyze the trade-offs between model performance and deployment constraints in real-world scenarios.
• "Hardware Selection Strategy for AI Systems" - Compare edge devices, cloud GPUs, and self-hosted solutions. Discuss cost-benefit analysis and provide recommendations for different business scenarios.

MVP Development & Business Applications

• "Rapid AI MVP Development: Lessons from Industry" - Analyze successful AI MVP case studies, discussing how the KISS principle and template-driven development accelerated time-to-market. Include failure cases where over-engineering delayed product launch.
• "From Prototype to Production: The Reality Gap" - Examine common challenges when transitioning AI models from development to production environments, including data drift, infrastructure scaling, and monitoring requirements.

Infrastructure & DevOps

• "Container Orchestration for ML: Docker vs. Kubernetes Trade-offs" - Analyze when containerization adds value versus complexity in ML projects. Discuss deployment strategies and infrastructure management challenges.
• "CI/CD for Machine Learning: Unique Challenges and Solutions" - Compare traditional software CI/CD with ML-specific requirements. Discuss automated testing strategies for models and data pipelines.

Presentation in the last lecture

Description:

Students will work in small groups (3-5 people) to develop a blueprint for a dummy artificial intelligence (AI) project covering all stages of the AI/ML lifecycle. Each student acts as project manager – the emphasis is on planning, analysis and communication skills rather than technical execution.

Project content:

• Description of the problem and business context
• as AI could help address this problem
• available or supposition data sources
• data acquisition and processing plan
• The type and suitability of the potential model
• key risks and limitations
• definition and measurement of a successful result

end result:

• Presentation (10-15 min) in the final lecture of the course

evaluation criteria:

• Logical and sound design structure
• problems and relevance of solution
• understanding the choice of data and models
• definition of risk awareness and measurements