A collection of not-so-memorable thoughts and musings.
Welcome! This blog documents real technical work from building production systems.
I’m working on Colosseum, a multi-agent investment committee framework that uses AI agents to collaboratively analyze markets and make trading decisions.
Recent Topics:
All posts are based on actual code, real design decisions, and lessons learned from building systems that work.
Check out my latest posts below, or visit the archive to browse all articles.
Last Thursday, Globalstar (GSAT) spiked 26% afterhours on rumors that Amazon was in talks to acquire the company. I wanted to build a trading game plan around it. What started as a simple exercise turned into a lesson in why data quality matters and ended with me publishing a set...
AI agents need memory. Without state persistence, every conversation starts from scratch, agents can’t coordinate across sessions, and you lose all audit trails. Here’s how I designed the database layer for Colosseum.
I’ve been building Colosseum, a multi-agent framework that simulates an investment committee using AI agents. Instead of a single AI making trading decisions, Colosseum orchestrates multiple specialized agents that collaborate like a real investment team.
Last week, I migrated Colosseum from Kubernetes (KIND) to Podman Quadlets. The result? Simpler deployment, better localhost networking, and native Linux integration. Here’s why and how I did it.
One of the biggest challenges in building AI trading agents is securely connecting them to broker APIs. How do you give an AI agent access to real money without creating a security nightmare?
LangChain’s initialize_agent() is deprecated. If you’re still using it, you’re building on a dead API. Here’s how I migrated Colosseum to modern LangGraph patterns and why you should too.