PRISM-INSIGHT

Hey IH,

I've been running PRISM-INSIGHT for about 10 months now, and I thought I'd share what started as a personal frustration project.

The problem was simple: I was wasting 2-3 hours every morning before market open just doing research. Reading charts, scanning news, digging through financials, checking insider activity. It's tedious, and you always feel like you're missing something.

So I built a system where I let AI handle the grunt work. Instead of one AI model trying to do everything, I broke it into 13 specialized agents - kind of like assembling a research team:

- Technical Analyst watches price patterns and volume

- Fundamental Analyst digs into financials and earnings

- News Analyst tracks sentiment from recent coverage

- Institutional Flow Analyst monitors what big money is doing

- And 9 others covering different angles

They work asynchronously, pass their findings to each other, then an Investment Strategist synthesizes everything into a recommendation. It's way more thorough than I could do manually.

I launched it in March 2025. It supports both Korean (KOSPI/KOSDAQ) and US (NYSE/NASDAQ) markets. I send alerts via Telegram - got 550+ subscribers now, mostly people I've never met. The system can actually auto-trade too if you connect a broker API.

Real talk on costs: running this costs about $300/month in API calls and server costs (OpenAI, Anthropic, various data APIs). That's the main constraint. On the revenue side, it's zero right now - it's open source under AGPL-3.0, no business model yet.

Landing page: https://prism-insight-landing.vercel.app/

GitHub: https://github.com/dragon1086/prism-insight

Live dashboard: https://analysis.stocksimulation.kr

Two things I'm genuinely curious about from this community:

1. Do you think the multi-agent approach actually adds value, or am I over-engineering it? Could a well-crafted single prompt do most of what these 13 agents do?

2. How do you handle API costs for side projects that don't generate revenue? I'm wondering if there's a smarter way to optimize without just turning everything off.

Would love to hear thoughts, especially from anyone working on AI projects with similar cost structures.