gameworkerkim/vibe-investing

What it solves

This repository provides a comprehensive ecosystem for "Vibe Investing," an agentic pipeline that uses LLMs to collect, analyze, and validate investment decisions based on market data, news, and on-chain signals. It aims to move beyond simple LLM prompts to create a structured, multi-model "investment committee" that reduces noise and identifies genuine market signals for stocks, crypto, and luxury sectors.

How it works

The project implements several specialized tools and frameworks:

• Multi-LLM Investment Committee: Uses different models (Claude, Gemini, ChatGPT, DeepSeek) as distinct personas (e.g., Risk Manager, Quant Strategist) to cross-verify macro diagnoses and execution strategies.
• CASSANDRA AI: A system that monitors DART electronic disclosures to detect structural risks in the KOSDAQ market (e.g., paper companies, suspicious M&As) using multi-LLM ensembles.
• ARDS-X Regime Classifier: A quant tool that uses FRED and yfinance data to automatically classify market states into five categories (Normal, Correction, Oversold, Distribution, Recession) to determine which trading strategy to apply.
• TokenForge: A prompt optimizer that converts Korean prompts into "caveman-ultra English" to reduce token consumption for coding agents.
• LAON VaultGuard: A security tool that uses LLMs to detect leaked private keys in Git repositories, supporting offline mode via Ollama.
• Toss x AMQS Dashboard: A full-stack dashboard applying momentum rules to Korean stocks and ETFs via the Toss Open API.

Who it’s for

• Quant traders and investors looking for AI-driven market analysis and risk management tools.
• LLM developers interested in prompt engineering for finance and token optimization.
• Financial analysts who want to automate the monitoring of corporate disclosures and relationship mapping.

Highlights

• Agentic Investment Pipeline: Shifts from static prompts to active tool-calling and multi-agent consensus.
• Diverse Toolset: Includes everything from regime classifiers and disclosure monitors to security scanners and token optimizers.
• Curated "Awesome" Lists: Provides deep evaluations of quant scripts, trading bots, and prompt libraries specifically for finance.
• Practical Backtesting: Includes real-world backtest reports for Korean and US markets (e.g., ARDS-X and AMQS).
• Multi-Model Ensemble: Leverages the strengths of different frontier LLMs to prevent single-model bias in financial decision-making.