December 30, 2025 · Build Log
Building AI-Powered Call Automation
When we decided to automate our call-recording workflow at work, I knew off-the-shelf solutions wouldn't cut it. We needed something that could transcribe calls, analyze sentiment, extract action items, and present it all in one dashboard.
The stack
React 18 on the frontend for its ecosystem and component reusability. FastAPI on the backend — it's fast, documents itself, and plays well with Python's ML libraries.
For transcription, OpenAI's Whisper was the obvious choice: accurate, quick, and multilingual. Transcripts then run through GPT-4 for sentiment analysis and action-item extraction.
What I'd tell anyone building the same thing
- Start with Docker. Containerizing from day one made deployment trivial.
- Real-time matters. People want live updates, not page refreshes.
- API design is crucial. A well-designed REST API makes the frontend almost build itself.
- Watch your costs. API calls add up fast — cache aggressively.
The system now processes hundreds of calls a day, categorizing them by sentiment and flagging urgent action items automatically. It has saved the team countless hours of manual review.
December 20, 2025 · Post-Mortem
From Employee to Entrepreneur
Starting Mountain West Surface while working full-time at Ski Butlers was one of the harder decisions I've made — and one of the most rewarding.
Why I started
Living in Utah's mountain communities, I kept noticing a gap in quality exterior care. Properties here face harsh winters and extreme weather, and the service side of the market hadn't caught up. I saw room for professional, reliable work with a systems-first approach.
Balancing both worlds
The first six months were brutal: full days at work, service calls on evenings and weekends, customers answered at night. What made it work:
- Systems over hustle. I automated everything I could — scheduling, invoicing, customer communications.
- Quality over quantity. Start small, do perfect work, and let word-of-mouth do the growing.
- Leverage what you know. A tech background meant I could build the website, run the marketing, and design efficient workflows myself.
The part I didn't expect
Running my own business made me better at my day job, and the day job — service, all day, for people with high expectations — made me better at running the business. The skills transfer in both directions.
If you're considering entrepreneurship while employed: start small, validate the market, build systems, and don't quit your day job until the numbers make sense. Your current job is your runway — use it wisely.
December 10, 2025 · Reality Check
Algorithmic Trading: Reality vs. Hype
After building multiple trading bots and running them with real money, I can report that algorithmic trading is nothing like what YouTube gurus make it seem.
The reality check
Most retail traders lose money — even with sophisticated algorithms. The market is brutally efficient, and any edge you think you've found has probably already been exploited by firms with more resources, faster execution, and smarter people.
What actually matters
- Edge is everything. Without a genuine statistical edge, you're gambling. Period.
- Risk management beats strategy. Position sizing and stops matter more than entry signals.
- Backtests lie. Every backtest overfits. Forward testing reveals the truth.
- Costs kill. Slippage and fees eat profits faster than you'd expect.
- Psychology still applies. Even with bots, watching your account swing is emotional.
My approach now
I use a relative-strength methodology: the system scans for setups against the broader market, and I keep the size small and the stops tight. Small wins, early exits, boring consistency.
Every trade the system takes is published on the live stats page — including the losing ones. If I'm going to write about trading honestly, the numbers should be public too.
Bottom line: algorithmic trading is a tool, not a money printer. Treat it like engineering — test rigorously, manage risk obsessively, and be honest about your edge.