Blog

Technical deep-dives, lessons learned, and thoughts on building production ML systems.

May 24, 202614 min

Extending AI Coding Assistants: Claude Code vs. GitHub Copilot vs. Cursor

Agents, skills, commands, instructions, hooks, plugins, connectors — seven primitives, three tools, one cross-vendor standard that quietly won. What's actually portable, what's marketing, and what to author once if you want it to run everywhere.

AI ToolsClaude CodeGitHub CopilotCursorMCP

May 20, 202611 min

Migrating a Production LLM Backend from AWS to DigitalOcean

What actually broke when we replatformed an LLM inference service off ECS Fargate — connection pooling, a missing env var, and a silent fallback that hid a broken model path.

InfrastructureMigrationFastAPIDigitalOcean

February 15, 20269 min

Cutting LLM Cost 95% Without Losing Accuracy

How we moved a high-volume classification workload from GPT-4 to a fine-tuned cheaper model — eval harness first, prompt work second, fine-tuning to close the gap.

LLMCost OptimizationFine-tuningEvals