A two-stage pipeline for taming quant research overload

What it does

QuantMind is an async Python framework that ingests unstructured financial documents—primarily arXiv papers today—and converts them into a structured knowledge graph. It splits the work into two stages: an extraction pipeline that parses PDFs and HTML, tags topics, and deduplicates content, and a retrieval layer that serves the results through embeddings, RAG, and multi-hop reasoning. The aim is to let quant teams explore factor strategies and risk models via natural language rather than manual review.

The interesting bit

The framework includes a magic resolver that converts free-form strings like “Pull arXiv 2401.12345 about cross-sectional momentum” into structured pipeline inputs, which saves researchers from wrangling configuration objects. More unusual is the README’s candor: the project is mid-migration to the OpenAI Agents SDK, with persistent memory and the store layer still scheduled for future PRs, and the sweeping vision of an all-encompassing financial intelligence layer is explicitly labeled as aspirational, not shipped.

Key highlights

• Decoupled two-stage architecture separating knowledge extraction from intelligent retrieval
• Async batch processing with concurrency controls and error handling via batch_run and paper_flow
• Domain-specific LLM tuning for finance content and automatic categorization into research topics
• Natural-language intent resolver (magic) that maps free-form requests to structured inputs
• Accepted at NeurIPS 2025 GenAI in Finance Workshop

Caveats

• The agent layer is mid-migration to OpenAI Agents SDK, and the mind/ memory and persistent store layers are not yet implemented
• Production examples currently center on arXiv papers; broader sources such as news, blogs, and SEC filings remain on the roadmap
• The README explicitly warns that the “intelligent research agent” vision describes long-term goals, not current capabilities

Verdict

Quant teams drowning in academic literature should evaluate the paper pipeline; developers looking for a finished, general-purpose financial knowledge platform should wait for the remaining roadmap items to land.