AlphaCouncil: Multi-Agent Debate-Driven Taiwan Stock Analysis with Google ADK + Open-Source LLMs — Localizing Both Data and Models
- Time
- 2026-08-09 14:00 ~ 14:30
- Speaker
- Samson, 陳庭宇
- Room
- AU
- Co-write
- https://hackmd.io/rJmFk4g8Mg
Abstract
In recent years, two high-profile open-source investment-analysis implementations have made the idea of using LLMs to analyze stocks accessible to a much broader developer community. However, systems of this kind are often designed around English-language data, U.S. markets, and American investment contexts. Bringing them to Taiwan requires more than replacing a ticker symbol: the data sources, market terminology, and decision-making workflow all need to be adapted.
This talk presents AlphaCouncil, a localized multi-agent stock-analysis system built with Google ADK (Agent Development Kit). It uses ADK's native SequentialAgent, ParallelAgent, and LoopAgent primitives to orchestrate a complete research workflow. Multiple specialist agents collect market information in parallel, investment-master personas provide different perspectives, and bullish and bearish researchers debate the evidence. A research manager then consolidates the arguments, a trader turns the decision into an execution plan, and a final risk debate informs the portfolio manager's decision.
The localization begins at the data layer. For the Taiwan market, AlphaCouncil integrates free public data from TWSE, TPEX, MOPS, and TAIFEX, including institutional flows, market positioning, monthly revenue, valuation indicators, options signals, and market-risk metrics. These sources are combined with Taiwanese financial news so that the system can produce analysis grounded in Taiwan's market context rather than simply translating a U.S.-centric workflow.
At the model layer, AlphaCouncil supports running replaceable open-source models locally through Ollama. Model choices can be adapted to hardware constraints, Traditional Chinese capabilities, tool-calling support, and deployment requirements. This local-first approach helps keep financial data on-premises while reducing dependence on commercial API costs; commercial models can also be used when appropriate.
The talk focuses on three engineering challenges: orchestrating multi-agent workflows with native Google ADK primitives, integrating free Taiwan-market data into structured agent inputs, and adapting replaceable open-source models to produce Traditional Chinese analysis with Taiwan-specific financial context. Through a live demonstration and backtesting results, we will also discuss how to evaluate an agent that consistently follows a predefined investment philosophy, even when that philosophy underperforms the market in a particular environment.
Whether you are building domain-specific agents with open-source models or localizing an English-oriented agent system for a Traditional Chinese audience, AlphaCouncil offers a decomposable, replaceable, and extensible reference implementation.
AlphaCouncil is intended for educational and research purposes only and does not constitute investment advice.
Speaker
Samson
現任國泰金控助理架構師，擁有 9 年以上系統架構與後端開發經驗，橫跨金融、政府、醫療與車用等多個產業領域。目前職責著重於企業 AI 賦能與架構治理，推動內部 AI 導入策略與工程實踐。
長期關注開源技術在垂直領域的落地應用，熟悉雲端部署與分散式系統設計。
本次議程以自行實作的 AlphaCouncil 開源專案為主軸，分享在金融場景中導入 Google ADK 與開源 LLM（Gemma 4）構建多代理人系統的工程實踐。
Assistant Architect at Cathay Financial Holdings with 9+ years of system architecture and backend development experience across financial services, government, healthcare, and automotive domains. Currently focused on enterprise AI adoption and architecture governance.
A practitioner interested in applying open-source technology to real-world vertical domains, with background in cloud deployment and distributed systems design.
This talk is grounded in hands-on experience building AlphaCouncil: an open-source multi-agent Taiwan stock analysis framework integrating Google ADK with locally-deployed open-source LLMs (Gemma 4).