Projects¶
Real experiments, real outcomes, real learnings. Each project follows a structured format to capture what worked, what didn't, and what's next.
Project Case Studies¶
Agentic Workflow Prototype v1¶
Context
After learning about agentic workflows through AI Week and various tool deep dives, I wanted to build a real workflow that could handle a multi-step task autonomously. The goal was to understand how to structure prompts, manage state, and evaluate outputs in a production-like context.
Problem
Manual research and synthesis tasks were taking significant time. I needed a system that could: - Gather information from multiple sources - Synthesize findings - Present results in a structured format - Handle errors gracefully
Approach
Built a prototype using n8n as the orchestration layer, with ChatGPT API for reasoning and synthesis. Structured the workflow with clear decision points, error handling, and output validation. Tested with real research questions and iterated on prompt design.
AI Tools Used
- n8n (workflow orchestration)
- ChatGPT API (reasoning and synthesis)
- Cursor (code development and debugging)
Outcome
Successfully built a working prototype that could handle multi-step research tasks. Learned critical lessons about: - The importance of clear prompt structure - How to design decision points in agentic workflows - The need for human-in-the-loop validation for critical outputs - The complexity of managing state across multiple steps
Next Iteration
- Add more robust error handling and retry logic
- Implement better output validation
- Test with more complex research questions
- Explore adding memory/context management
- Document the workflow as a reusable template
AI Week Learning Playbook¶
Context
During AI Week, I was exposed to a wide range of AI tools, concepts, and approaches. I needed a way to synthesize this learning and create a structured path forward for my AI PM journey.
Problem
Information overload. Too many tools, too many concepts, not enough structure. I needed a learning system that would help me: - Organize information - Create a clear learning path - Track progress - Apply learnings to real projects
Approach
Created a structured playbook using Markdown, organized by topic areas (tools, concepts, workflows). Built it iteratively, adding sections as I learned. Used AI tools to help structure and refine content, but maintained control over the organization and priorities.
AI Tools Used
- ChatGPT (content structuring and refinement)
- Cursor (documentation and organization)
- Napkin (note-taking and synthesis)
Outcome
A comprehensive learning playbook that serves as both a reference and a roadmap. It's living documentation that evolves as I learn. The process of creating it helped me synthesize information and identify gaps in my understanding.
Next Iteration
- Add more tool-specific deep dives
- Create project templates based on learnings
- Build evaluation frameworks for AI outputs
- Document common patterns and anti-patterns
This Website (LukeMoody.ai)¶
Context
I needed a platform to document my journey, share experiments, and build in public. Wanted something that felt personal, tech-forward, and clearly built with AI collaboration—not a templated Squarespace site.
Problem
Most website builders are either too templated or too complex. I needed: - Full control over structure and styling - Easy content updates - A way to showcase the "built with AI" philosophy - Professional but personal feel
Approach
Chose MkDocs + Material for its balance of simplicity and customization. Designed the structure collaboratively with AI tools (Atlas + Leo), then built it manually. Created custom CSS for terminal-style elements and AI stack visualization. Wrote all content myself, using AI for structure suggestions and refinement.
AI Tools Used
- Atlas + Leo (initial structure and design concepts)
- ChatGPT (content refinement and suggestions)
- Cursor (code development and styling)
Outcome
A fully functional, launchable website that reflects my voice and philosophy. The process demonstrated the "built with AI" approach: AI for ideation and structure, me for execution and decisions. The site itself is a case study in AI collaboration.
Next Iteration
- Add more project case studies as I build them
- Implement analytics to understand what resonates
- Add RSS feed for updates
- Consider adding a blog section for shorter-form updates
Luke's AI Playground¶
Small experiments, quick iterations, "dumb but shipped" projects. These are the tiny tests that inform bigger work.
Agentic Browsing Test¶
What: Tested using AI agents to browse and extract information from websites autonomously.
Tools: Lindy, custom browser automation
Learnings: Agents can navigate simple sites well, but struggle with complex interactions. Need clear success criteria.
Status: Shipped, documented, moving on
n8n Draft Workflow¶
What: Built a workflow that takes a topic and generates a structured draft document using multiple AI steps.
Tools: n8n, ChatGPT API
Learnings: Multi-step workflows need careful state management. Error handling is critical.
Status: Working prototype, iterating
Napkin Notes Pipeline¶
What: Automated the process of taking notes in Napkin and exporting them to structured Markdown.
Tools: Napkin, n8n, custom scripts
Learnings: Automation works best when it fits existing workflows, not replaces them.
Status: In progress
LM Studio Experiment¶
What: Tested running local LLMs for privacy-sensitive tasks.
Tools: LM Studio, local models
Learnings: Local models are slower but useful for sensitive data. Trade-offs are real.
Status: Experiment complete, documented
Prompt Library System¶
What: Built a system to store, version, and test prompts for different use cases.
Tools: Markdown files, custom evaluation scripts
Learnings: Prompt versioning is as important as code versioning. Testing frameworks help.
Status: Early stage, iterating
Project Principles¶
Every project I take on follows these principles:
- Ship Fast, Iterate Often — Get something working, then improve it.
- Document Everything — What worked, what didn't, and why.
- Real Use Cases — Build things I actually need, not just demos.
- Learn Publicly — Share failures and successes.
- AI as Partner — Use AI to amplify, not replace, my thinking.
If you're working on similar projects or want to collaborate, let's talk.