AI/LLM Engineering
Pioneering AI-assisted development methodologies and building enterprise-scale applications with LLM integration.
Learning to Pair With LLMs
I was working with LLMs before it was cool—literally. While building Discussit in 2023, I was already integrating OpenAI's APIs for transcription, conversation analysis and intelligent summarization, experimenting with token-aware chunking strategies and hierarchical content processing.
The Dream100 Revelation
Building the Dream100 platform changed everything. The technical challenge was straightforward—scrape YouTube channels, generate embeddings, match content using vector similarity search. But something remarkable happened during development: I built the entire system in 20 days using Python and FastAPI, frameworks I barely knew, by effectively communicating with an LLM.
This wasn't about prompting tricks or finding the perfect model. The revelation was architectural. The AI excelled when I stopped thinking about implementation details and started thinking about system design.
The Great Todo App Experiments
Experiment 1: Rust
Discovered the "bootstrap phase" challenge—initial setup where LLMs struggle with project structure and dependency management.
Experiment 2: Meta-Level Thinking
Created comprehensive architectural documentation upfront, dramatically improving consistency.
Experiment 3: Documentation Synthesis
Pioneered using LLMs to create custom architectural guides by combining multiple sources.
Experiment 4: C# Discovery
Found that C#'s verbose nature became a strength in AI-assisted development.
Real-World Applications
PowerPort
Started as an application to help streamline PowerBI deployment and management, and evolved into a data aggregation platform.
AstralMCP
Sophisticated middleware system enabling AI models to interact with REST APIs through dynamic MCP tool generation.
The Meta-Lesson
"The more ambitious your goals, the higher the level of abstraction you need to operate at. This shift—from implementation thinking to architectural thinking—fundamentally changed how I approach software development."
Current State and Future
Through Generate Labs, I've established repeatable patterns for AI-assisted development that reduce development time by 60-80% while maintaining architectural integrity.
The future of software development isn't about replacing developers with AI. It's about developers learning to think architecturally while AI handles implementation. Those who master this partnership will build things that seemed impossible just a few years ago.
Key Learnings
LLMs excel at architectural guidance which provides structure for implementation
Language choice significantly impacts AI-assisted development efficiency
Documentation synthesis is crucial for maintaining consistency
Meta-level thinking produces better results than implementation-level prompting