Multi-agent AI stock analyzer – 408% return trading Korean market
Hey HN! I built PRISM-INSIGHT, a multi-agent system where 13 specialized AI agents collaborate to analyze Korean stocks (KOSPI/KOSDAQ). It's completely open source and has been running live since March 2025. [What it does] The system automatically detects surging stocks twice daily, generates analyst-level reports, and executes trading strategies. Each agent specializes in something different – technical analysis, trading flows, financials, news, market conditions, etc. They work together like a real research team. [Why I built this] I wanted to see if GPT-4 and GPT-5 could genuinely…
What it does
In the maker’s words, at launch
Hey HN! I built PRISM-INSIGHT, a multi-agent system where 13 specialized AI agents collaborate to analyze Korean stocks (KOSPI/KOSDAQ). It's completely open source and has been running live since March 2025. [What it does] The system automatically detects surging stocks twice daily, generates analyst-level reports, and executes trading strategies. Each agent specializes in something different – technical analysis, trading flows, financials, news, market conditions, etc. They work together like a real research team. [Why I built this] I wanted to see if GPT-4 and GPT-5 could genuinely replicate what human analysts do, but without the typical single-agent limitations. So I split the work across multiple specialized agents that collaborate. The trading simulation has been running for 8 months now with real Korean market data. [How to try it] Join the live Telegram channel! https://t.me/prism_insight_global_en (gets daily alerts and reports) Check the real-time dashboard! https://analysis.stocksimulation.kr (all trades, performance, AI reasoning) Clone and run it yourself! https://github.com/dragon1086/prism-insight [The interesting parts] The system uses MCP (Model Context Protocol) servers to give agents access to live market data, web search, and financial APIs. I'm using GPT-4.1 for analysis, GPT-5 for trading decisions, and Claude Sonnet 4.5 for the conversational bot. The first trading simulation (Season 1, Mar-Sep 2025) returned 408% across 51 trades. Current season(2) is at +11% realized returns vs KOSPI's +16%. Also running it with real money now ($10k account, up 9.35% since late September). [Tech stack] Python 3.10+, async/await throughout, SQLite for trade history, Playwright for PDF reports, matplotlib for charts. The whole thing is about 8,400 lines of Python across 56 files. [What makes it different] Most AI trading projects are either single-agent or black boxes. This one uses a multi-agent architecture where you can see exactly what each agent is analyzing and why. Everything is transparent – the dashboard shows every trade, every decision, and all the reasoning. It's MIT licensed and runs entirely on your machine if you want. I'm covering the API costs (~$200/month) to keep the public Telegram channel free for 450+ users(Korean channel + Global channel). Would love feedback on the multi-agent approach or questions about running AI agents in production!
Does the same job
all alternatives →
- IMI made Quantify AI – alternative to TradingView (2 months work)Oct 2025 · quantify-ai.co · ▲6
How it works (tech stack): -Built entirely with Lovabl.dev (no-code front-end + logic) -ChatGPT / Claude for research and inspiration -Powered by GPT-4 Vision to interpret charts visually -Hosted on Supabase for performance & caching It’s not meant to replace analysts — just to speed up how traders interpret data. I’m a designer exploring AI tools, and this is my first attempt to turn an idea into a functional product. Would love to know what you think.
- FFFramework for building multi-agent equity research agentsFeb 2026 · github.com · ▲6
I built Hermes, an open-source Python framework for multi-agent financial research. Most AI “equity research” demos stop at generating text. In practice, real workflows require pulling structured XBRL financials from SEC filings, extracting labeled sections like MD&A and Risk Factors, merging macro and market data, building actual Excel models with formulas, and generating investment memos in Word or PDF. Hermes is designed to handle that full pipeline end to end. It includes 35 financial data tools covering SEC EDGAR (via edgartools), FRED, Yahoo Finance market data, and RSS-based financial…
- 1R13Radar – Real-Time Hedge Fund Portfolio AnalyticsNov 2025 · 13radar.com · ▲7
Hi HN, We’re a small team working on 13Radar.com, which we launched about two weeks ago after 4 months of development. I’m the founder, and together with the team we’re building a platform that tracks hedge fund portfolios in real-time based on SEC Form 13F filings. AI has been a major helper in our workflow. For a single webpage, we often consult multiple AI systems in parallel, generating different versions and comparing them side by side before deciding on the final design or implementation. More than 60% of the research, design, and coding involved AI assistance. For UI design we used…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, November 2025
the whole month →
- IB
Life & fun · Nov 2025 · bitsnpieces.dev
- BBoing▲782
Life & fun · Nov 2025 · boing.greg.technology