I built an AI to analyze stocks after losing money trading

1. Why I built this

I kept losing money trading U.S. stocks, and realized most of my time was spent reading news and trying to interpret it.

After a while, it became repetitive and exhausting.

So I built a small tool for myself.

1. What it does

It takes in news and SEC filings, and turns them into structured daily summaries.

Instead of reading dozens of articles, I can quickly see:

• what moved the market
• key takeaways per ticker
• overall market flow after the close

3.Key features

• Daily market close reports
• On-demand per-ticker analysis (cached after first run)
• ETF composition and trend insights

1. Challenges

One of the hardest parts was balancing cost vs performance.

Initially, I wanted real-time generation, but LLM + data API costs were too high.

So I switched to:
→ generate once after market close

→ cache results for fast access

This reduced costs significantly.

1. What I learned

• AI feels more like amplification than replacement
• Small problems can take an entire day to solve
• Infrastructure (AWS, CI/CD) was harder than expected

1. Open question

Even with sufficient ECS + RDS resources, the first request feels slow.

Subsequent requests are fast (due to caching), but initial load is delayed.

My guess is:
→ cache miss → generation pipeline → LLM latency

Curious if anyone has dealt with similar issues.

1. Link

If you're curious, here's what I built:
https://rallypi.com