
One AI Parameter Change Cost Me $54/Month
I thought my Cloud Run migration was flawless until a single AI parameter—temperature set to 0.7 instead of 0—caused 30% failed API calls and cost me $54/month in wasted tokens.
80 posts in this category

I thought my Cloud Run migration was flawless until a single AI parameter—temperature set to 0.7 instead of 0—caused 30% failed API calls and cost me $54/month in wasted tokens.

I migrated from AWS to GCP and achieved 92% cost reduction with 10x faster performance—here's what I learned ditching "best practices" for pragmatic architecture that actually works.

I built an AI podcast generator that lets you shape the content before it's created—because I was tired of AI just reading Wikipedia at me during my commute.

After 5,000 hours testing AI models, Gemini 2.5 Pro beats Claude and ChatGPT for coding—here's why it became my default tool for building complex applications.

After 10 days building AI-generated podcasts with CrewAI, I learned that while the framework is remarkably easy to use, quality content still demands real human effort.

I scaled Sydney to read 10 years of narrative content from Big Tech's annual reports by strategically limiting scope to 7 companies—here's how I balanced cost vs capability.

I combined two AI agents into one multi-talented assistant that can analyze S&P 500 financials AND my 15 years of blog content—proving less really is more.

I built an AI agent that answers financial questions using 10 years of SEC data—and finally solved my streaming challenges to make it real-time and verifiable.

After 10 months of coding with AI assistants, I've learned they're like interns—powerful but needing specific instructions, which means I must deeply understand frameworks to solve real problems like making my slow chatbot fast.

I upgraded my chatbot Sydney to test Weaviate's hybrid search and query structuring—features I need to scale my financial chatbot across 500+ companies.

I built a financial chatbot that answers S&P 500 questions using SEC data—with a self-critique agent that improves answers before you see them.

I rebuilt my site's frontend with AI, upgraded my chatbot's intelligence, and discovered continuous deployment—all because Docker refused to cooperate.