Lab Notebook
Active experiments, research logs, and things I'm investigating. Each entry is a real experiment with observations and takeaways.
> experiments.active = 5
> experiments.completed = 2
> total_entries = 7
>
Smoothies — Fresh Pavement Finder
An Android app that scrapes city paving data to find freshly paved streets for longboarding. Because nothing beats riding on fresh asphalt. Pulls municipal road data, maps it, and alerts you to smooth new pavement in your area.
Municipal data is surprisingly accessible. The real challenge is normalizing it across different city formats and keeping it fresh.
LLM Prompt Engineering Patterns
Exploring advanced prompt engineering techniques including chain-of-thought, few-shot learning, and system prompt architecture for reliable AI outputs.
Structured prompts with clear constraints produce 3x more consistent outputs than freeform instructions.
n8n Automation Workflows
Building complex multi-step automation workflows using n8n to connect business tools, APIs, and AI services without traditional coding.
Visual workflow builders can handle 90% of business automation needs when combined with custom API nodes.
React Component Architecture
Studying modern React patterns including compound components, render props, custom hooks, and state management approaches for scalable front-end applications.
Custom hooks are the most powerful pattern for separating concerns in React applications.
AI Agent Frameworks
Experimenting with autonomous AI agent frameworks like LangChain and CrewAI to build multi-step reasoning systems that can decompose and solve complex problems.
Agent orchestration is the key to unlocking AI's potential beyond simple chat interactions.
Python Data Pipeline Design
Building robust ETL pipelines in Python for transforming messy business data into clean, analyzable datasets for decision-making.
Data validation at entry points prevents 90% of downstream pipeline failures.
RAG Knowledge Systems
Building Retrieval Augmented Generation systems that combine organizational knowledge bases with LLMs for context-aware AI assistants.
Chunking strategy and embedding model choice have more impact on RAG quality than the LLM used.