Navigating a Fast-Moving Field
A Year in Review: What I've Been Up To (And What I've Learned)
Reflections on growth, working in AI, and walking the dog.
Writing
AI in Practice
Improving Agent Systems & AI Reasoning
DeepSeek-R1, OpenAI o1 and o3, test-time compute scaling, model post-training — and what the shift toward Reasoning Language Models actually means for the people building agent systems on top of them.
AI in Practice
Overcome Failing Document Ingestion & RAG Strategies with Agentic Knowledge Distillation
This piece introduces the Pyramid Approach, an agentic knowledge distillation technique that transforms messy source documents into structured, retrieval-ready knowledge.
Research
Introducing Layer Enhanced Classification (LEC)
A novel approach to lightweight safety classification that outperforms GPT-4o on content safety and prompt injection detection — using fewer than 100 training examples and a 0.5B parameter model. Here's how it works and why it matters.
AI in Practice
Computer Use and AI Agents: A New Paradigm for Screen Interaction
When AI agents can see and interact with a screen the way a human does, what actually changes? This piece explores the multimodal shift and what it means for how we build and deploy agent systems.
AI in Practice
AI Agents: The Intersection of Tool Calling and Reasoning in Generative AI
Tool calling is often described as magic. It isn't. This piece unpacks exactly how function calling works under the hood — and why the interplay between tool use and model reasoning is what makes or breaks agentic systems.
Navigating a Fast-Moving Field
Navigating the Latest GenAI Model Announcements — July 2024
GPT-4o mini, Llama 3.1, Mistral NeMo 12B — July 2024 brought a wave of model releases. Here's a clear-eyed guide to what actually changed, what's worth paying attention to, and what's mostly noise.
AI in Practice
Understanding Techniques for Solving GenAI Challenges
Pre-training, fine-tuning, RAG, prompt engineering — these aren't just buzzwords. This piece breaks down the actual mechanics of each approach and helps you choose the right technique for your specific problem.
AI in Practice
Are Language Models Benchmark Savants or Real-World Problem Solvers?
Leaderboard scores tell you how a model performs on a benchmark. They tell you much less about how it performs on your problem. This piece examines the gap between evaluation and deployment — and what it means for how we assess AI progress.