Applying Large AI Models to Financial Data Governance and Innovative Use Cases

The presentation begins with an analysis of the current state of financial data production, covering regulatory guidelines, industry trends, and the traditional pipeline that relies heavily on web crawling, parsing, NLP extraction, and manual quality control.

It then explores the transition from AI‑assisted to AI‑native data production, highlighting changes in input/output formats, natural‑language interaction, and prompt‑engineering that shift the workflow from human‑centric to model‑centric processes.

Next, the article examines AI applications in data governance, describing the challenges of partial focus, neglect of end‑to‑end value, and insufficient security, and proposes a governance architecture that integrates APIs, private‑domain, and edge models with monitoring and risk‑control layers.

The concept of a lakehouse ("Lake‑Warehouse") as an AI‑friendly storage model is introduced, detailing how raw, structured, and unstructured data are unified, vectorized, and served to downstream AI services.

Enterprise‑level private‑domain large models are discussed, emphasizing compliance, interpretability, and consistency, and outlining three application categories: AI‑driven virtual assistants, productivity tools, and risk‑control systems.

A data‑centric versus model‑centric paradigm comparison is provided, followed by concrete data‑engineering analysis methods such as expression capability, data‑service provision (NL2API, NLP2SQL), and prompt‑engineering with domain experts.

Finally, the multi‑agent framework for intelligent investment advisory is described, showing how agents orchestrate RAG‑enhanced retrieval, expert modules, and risk‑control components to deliver high‑value, explainable financial recommendations.