Skip to content

Course Description

Overview

AI Problem Framing for AI Practitioners is a graduate-level course that teaches the strategic thinking skills needed to make high-stakes decisions about AI projects before committing to implementation. While most AI education focuses on building and optimizing models, this course addresses the upstream question: What should we build in the first place?

The course centers on a fundamental tension in applied AI work: practitioners face ambiguous business problems ("improve customer engagement," "reduce operational costs," "personalize learning") but must translate these into specific technical implementations. The gap between business ambiguity and technical specificity is where most AI projects fail—not due to poor model performance, but due to solving the wrong problem or choosing inappropriate solution approaches.

Through systematic frameworks and real-world case studies, you'll learn to deconstruct vague objectives into actionable opportunities, evaluate diverse AI solution alternatives, establish meaningful success metrics, and diagnose when projects need strategic pivots. This course teaches you to think like a strategic advisor who happens to have deep AI expertise, rather than a technician waiting for well-specified requirements.

The curriculum is built around The Loop framework—a repeatable process for moving from outcome clarity through problem deconstruction, alternative generation, trade-off analysis, and signal-based decision making. You'll practice applying this framework to cases spanning healthcare, education, finance, and enterprise operations, building a portfolio that demonstrates your ability to frame problems systematically before writing code.

Learning Outcomes

After completing this course, you will be able to:

  1. Analyze ambiguous business objectives to extract specific, measurable outcomes that can guide AI solution development
  2. Deconstruct complex problems into component parts, identifying constraints, stakeholders, success criteria, and hidden assumptions
  3. Generate diverse AI solution alternatives by systematically applying 13 solution archetypes (RAG, RLHF, fine-tuning, prompt engineering, etc.) to problem contexts
  4. Evaluate trade-offs between solution alternatives across dimensions of cost, latency, accuracy, interpretability, maintenance burden, and organizational fit
  5. Design signal frameworks that enable early detection of solution-problem mismatches through user behavior, system metrics, and qualitative feedback
  6. Interpret weak signals from live systems to distinguish meaningful patterns from noise and identify leading indicators of success or failure
  7. Decide when to persist with current approaches, pivot to alternative solutions, or stop AI initiatives based on systematic evidence evaluation
  8. Justify strategic recommendations about AI initiatives to technical and non-technical stakeholders using structured frameworks
  9. Critique existing AI problem framings to identify unstated assumptions, overlooked alternatives, and missing success criteria
  10. Synthesize complete problem framing analyses for novel AI opportunities, demonstrating the full Loop workflow from initial ambiguity to actionable implementation plans

Target Audience

This course is designed for:

  • AI/ML practitioners with production experience who want to make better strategic decisions about what to build
  • AI product managers who need frameworks for evaluating proposals and guiding technical teams
  • Technical leads responsible for scoping AI initiatives and allocating engineering resources
  • Graduate students in AI/ML programs preparing for applied research or industry roles
  • Researchers transitioning from academic to applied settings where business context shapes problem definitions

The ideal student has built at least one AI system from conception to deployment and has experienced the frustration of technically successful projects that fail to deliver business value. You should be comfortable with ML fundamentals and programming, but the course focuses on strategic thinking rather than implementation.

This course is not appropriate for:

  • Beginners without prior ML experience (take foundational coursework first)
  • Those seeking advanced modeling techniques (focus is on problem framing, not model optimization)
  • Practitioners looking for domain-specific solutions (course teaches general frameworks applicable across domains)

Prerequisites

Required Background

  • Machine Learning Fundamentals: Understanding of supervised learning, model training, evaluation metrics, overfitting, and generalization
  • Implementation Experience: Have built and deployed at least one ML system (academic project, internship, or production system)
  • Programming: Proficiency in Python and familiarity with standard ML libraries
  • Linear Algebra and Statistics: Comfort with concepts like probability distributions, statistical significance, and basic hypothesis testing
  • Experience with LLMs and prompt engineering
  • Exposure to production ML systems and their operational challenges
  • Understanding of software engineering practices (version control, testing, deployment)
  • Familiarity with business metrics and organizational dynamics

Assessment

This course uses portfolio-based assessment rather than traditional exams. Your work will demonstrate applied problem framing skills through progressively complex assignments:

Portfolio Components

  1. Framework Application Exercises (40%)
  2. Apply The Loop framework to 6-8 case scenarios
  3. Demonstrate systematic deconstruction, alternative generation, and trade-off analysis
  4. Each exercise builds specific skills (outcome clarity, constraint identification, signal design)

  5. Evaluated on depth of analysis and systematic application of frameworks

  6. Case Study Analyses (30%)

  7. Written analyses of 3-4 real-world AI projects
  8. Diagnose what went well, what failed, and why
  9. Propose alternative framings that could have led to better outcomes

  10. Evaluated on critical thinking and evidence-based reasoning

  11. Capstone Problem Framing (30%)

  12. Complete problem framing for a real or realistic AI opportunity
  13. Apply full Loop workflow from initial ambiguity to actionable implementation plan
  14. Include outcome specification, alternative evaluation, trade-off analysis, and signal framework
  15. Evaluated on comprehensiveness, rigor, and practical viability

Peer Feedback (Optional)

While not required for completion, peer feedback is strongly encouraged. Reviewing others' problem framings helps you:

  • Recognize patterns across different framings of similar problems
  • Develop critical evaluation skills by identifying gaps in others' analyses
  • See alternative approaches you might not have considered
  • Practice giving constructive feedback on strategic thinking

Grading Philosophy

Portfolio assignments are evaluated holistically based on:

  • Systematic application of frameworks (not just intuitive reasoning)
  • Depth of analysis (exploring second-order effects, hidden assumptions, edge cases)
  • Evidence-based reasoning (grounding claims in data, examples, or case study evidence)
  • Clarity of communication (structured arguments, clear trade-off articulation)

There are no "right answers" in problem framing—only well-reasoned or poorly-reasoned framings. Strong work demonstrates systematic thinking, considers multiple perspectives, and acknowledges uncertainty where appropriate.

Course Schedule

This course is self-paced and designed for completion over 6-8 weeks with 5-7 hours per week, totaling approximately 40 hours.

Chapter 1: The AI Problem Framing Mindset (6 hours)

  • Why AI projects fail at the framing stage
  • The solution-first trap and how to avoid it
  • Intellectual humility and working with ambiguity
  • Outcome clarity vs. solution specification
  • Assignment: Analyze 2 failed AI projects through framing lens

Chapter 2: AI Solution Alternatives (8 hours)

  • The 13 AI solution archetypes (RAG, RLHF, fine-tuning, prompt engineering, hybrid systems, etc.)
  • Matching archetypes to problem characteristics
  • Common anti-patterns and premature convergence
  • Building a mental model of the solution space
  • Assignment: Generate 5+ alternatives for 3 problem scenarios

Chapter 3: The Loop Framework (7 hours)

  • The complete Loop: Outcome → Deconstruction → Alternatives → Trade-offs → Signals
  • Systematic problem deconstruction techniques
  • Trade-off analysis across cost, latency, accuracy, interpretability, and organizational fit
  • Establishing meaningful success metrics and signal frameworks
  • Assignment: Apply full Loop to 2 case studies

Chapter 4: Diagnosis - Reading Signals (6 hours)

  • Weak vs. strong signals in AI systems
  • User behavior signals: adoption, engagement, workarounds
  • System performance signals: accuracy drift, edge case failures, latency patterns
  • Organizational signals: stakeholder feedback, resource constraints, competing priorities
  • Assignment: Design signal frameworks for 3 AI initiatives

Chapter 5: Pivot - Acting on Signals (6 hours)

  • The persist/pivot/stop decision framework
  • When to pivot vs. when to iterate
  • Types of pivots: problem pivot, solution pivot, outcome pivot, scope pivot
  • Managing sunk cost fallacy and stakeholder communication
  • Assignment: Evaluate 3 case studies and recommend persist/pivot/stop with justification

Chapter 6: Application - Full Case Studies (7 hours)

  • Healthcare: Clinical decision support system framing
  • Education: Personalized learning platform framing
  • Finance: Fraud detection system framing
  • Enterprise: Knowledge management system framing
  • Capstone Assignment: Complete problem framing for novel AI opportunity
  • Weeks 1-2: Chapters 1-2 (develop mindset, learn solution space)
  • Weeks 3-4: Chapter 3 (master The Loop framework)
  • Weeks 5-6: Chapters 4-5 (diagnosis and pivoting)
  • Weeks 7-8: Chapter 6 (synthesis through case studies and capstone)

You can accelerate or decelerate based on your schedule, but we recommend spending at least one week on Chapter 3 (The Loop) as it forms the foundation for subsequent work.

Course Philosophy

This course is built on several core beliefs:

Problem framing is a skill, not intuition. While some practitioners develop good instincts through experience, systematic frameworks enable more consistent and teachable problem framing.

The best solution to the wrong problem is still wrong. Technical excellence doesn't compensate for strategic misalignment. Learning when not to build AI is as valuable as learning how to build it well.

Ambiguity is the norm, not the exception. Real-world AI problems start messy. The ability to work systematically with ambiguity separates effective practitioners from those who need perfectly-specified requirements.

Pivoting is a feature, not a bug. The best AI teams pivot frequently based on evidence. This course teaches you to pivot strategically rather than randomly, and to recognize pivot signals early.

Context matters more than best practices. There is no universally best AI solution archetype. Everything depends on constraints, stakeholders, resources, and organizational context.

Instructor

Rajiv Shah, PhD Rajiv received his PhD from the University of Illinois Urbana-Champaign and currently works at Contextual AI. His research and industry experience span LLM alignment, production ML systems, and the organizational dynamics of AI adoption. He has advised dozens of AI initiatives across healthcare, education, and enterprise domains, with a focus on the strategic decisions that determine whether AI projects deliver measurable value.

Ready to begin? Start with the course home page or jump directly to Chapter 1: The AI Problem Framing Mindset.