At the ICLR 2025 Workshop on Advances in Financial AI, Jacob Chanyeol Choi and Joo Lee delivered an invited talk titled Compound AI Systems for Hedge Funds: Evolving from Single AI Models to Integrated Architectures. Their session explored how hedge funds can transition from standalone models to multi‑agent pipelines that integrate retrieval, analysis, and synthesis using LLM agents alongside quantitative frameworks. Key takeaways included deployment best practices, performance metrics for live systems, and lessons from real-world implementations. See session details → https://iclr.cc/virtual/2025/10000151

Alongside the talk, we unveiled FinDER, a real‑world Retrieval‑Augmented Generation (RAG) dataset for finance. FinDER contains 5,703 expert‑verified query–evidence–answer triplets sourced from SEC filings and annotated by former Goldman Sachs bankers. Unlike synthetic benchmarks, it challenges models to retrieve answers from unstructured documents using queries phrased as professional analysts would.

FinDER will be publicly released at ICLR ’25 with baseline evaluations and a leaderboard.

Both the Invited Talk and FinDER announcement underscore LinqAlpha’s dedication to advancing financial AI—from rigorous, expert‑annotated benchmarks to integrated multi‑agent architectures in high‑stakes environments.

If you are interested in LinqAlpha's Research, reach out at support@linqalpha.com.