As artificial intelligence technology rapidly reshapes the financial industry, a transformation in the investment research ecosystem, driven by large models, is accelerating and moving towards a new stage of deep intelligence. Facing industry pain points such as the underutilization of vast amounts of unstructured data like research reports and meeting minutes, and the difficulty of reusing knowledge, Ping An Fund has taken a leading step by officially launching its self-developed "AI Qing Fu"—an intelligent knowledge engine deeply tailored for fund investment research scenarios.
This platform not only reduces research report analysis time from several hours to minutes but also restructures the investment research workflow through intelligent tagging, cross-team sharing, and human-machine collaboration mechanisms. This lowers the cost of knowledge transfer, marking a new stage of self-sufficient, in-depth AI application for the company. In the race where leading institutions are actively deploying intelligent investment research systems, Ping An Fund is leveraging its proprietary technology as a fulcrum to drive the intelligent upgrade of its research framework and accelerate its bid to lead the AI-driven investment research era.
Timely Emergence: From Solving the "Last Mile" Problem to Building an Intelligent Knowledge Hub Currently, leading global financial institutions are accelerating their deployment of large model-driven intelligent investment research systems. However, despite rapid iterations in generative AI technology, the field of investment research still faces core bottlenecks such as low utilization of unstructured data, difficulty in knowledge reuse, and high costs for cross-team collaboration.
On one hand, the tens of thousands of daily new research reports and minutes contain significant value but remain largely dormant due to a lack of efficient processing tools. On the other hand, traditional general-purpose AI tools often focus on standardized scenarios like customer service and Q&A, struggling to meet the stringent demands of investment researchers for accuracy, traceability, and professional depth.
Against this backdrop, Ping An Fund's initiation of the "AI Qing Fu" project represents not just a technological upgrade but a strategic move for the future. The platform uses RAG technology as its core architecture to build a "data-knowledge-application" closed loop. It first employs Natural Language Processing (NLP) to intelligently segment, denoise, and perform semantic parsing on multi-source unstructured texts, including PDFs, Word documents, and transcribed meeting recordings, extracting key entities such as company names, industry classifications, and timelines along with their relationships. It then uses large language models for high-precision content annotation and converts text into high-dimensional semantic vectors for efficient storage and rapid retrieval.
When a user poses a question in natural language, the platform automatically performs semantic expansion to generate multi-dimensional search keywords. Combining dense vector retrieval with keyword matching, it quickly locates the most relevant knowledge fragments. A ranking model then optimizes result relevance, ultimately generating a structured summary of investment insights complete with source citations, covering logical chains, data support, and potential risk alerts.
This process fundamentally transforms the inefficient traditional model reliant on manual searching and data integration, reducing the time to extract core viewpoints from research reports from 30-60 minutes to under 5 minutes. The volume of information processed per unit of time has increased by over 90%, achieving true "second-level response and precise output."
Since the platform's launch, information processing efficiency has significantly improved, and barriers to cross-departmental knowledge sharing have gradually diminished. Investment research teams can now devote more energy to core activities like strategy design and in-depth analysis, enabling a more efficient allocation of resources towards high-value-added work. This efficiency gain not only streamlines internal communication but also helps the company progressively build an intelligent, shareable knowledge management system, transforming knowledge into a reusable asset.
More profoundly, beyond investment research, the influence of "AI Qing Fu" has permeated various dimensions of the company's operations. Its knowledge base capabilities have been extended to non-research scenarios such as human resources, financial compliance, and administration, spawning intelligent assistants like "HR Xiaolingtong." These can provide instant responses to standardized service queries regarding travel policies, IT issues, and reimbursement procedures, significantly enhancing cross-departmental collaboration efficiency. This intelligent operational model not only reduces comprehensive management costs but also further promotes the platform-based accumulation and immediate sharing of internal knowledge assets, building a smart organizational ecosystem where "everyone can search, and knowledge is usable everywhere."
Steady Progress: Building a Secure Foundation and an Evolutionary Path for Intelligent Investment Research As an AI platform applied in professional financial scenarios, security and reliability are paramount. To address potential "hallucination" issues in large models, Ping An Fund has innovatively established a dual-track control mechanism combining "technical constraints and manual verification." Technically, by lowering the model's temperature parameter and mandating that all generated content cite original sources with direct document links, it ensures every suggestion is traceable and verifiable. Procedurally, a "human-machine collaboration" mechanism requires that key investment recommendations must be manually reviewed and approved by researchers for reasonableness before adoption, achieving a deep integration of algorithmic efficiency and human experience.
Simultaneously, to guard against data security and privacy leakage risks, the platform has established a three-tier protection system of "input filtering, output purification, and behavior monitoring." The input side automatically identifies and blocks queries involving personal privacy or sensitive fields. The output side incorporates a dynamic keyword interception mechanism to prevent the generation of non-compliant content. The backend maintains complete logs of all user queries, intermediate reasoning, and final outputs, supporting full-process auditing and traceability. Additionally, Ping An Fund has implemented a quality assessment mechanism combining automated monitoring and manual random checks to continuously optimize platform performance.
Looking ahead, "AI Qing Fu" will continue to expand its application breadth and technical depth. On one hand, the platform will accelerate its rollout across the entire company, covering more business lines, and actively participate in industry exchanges to share replicable digital transformation experiences. On the other hand, the technical team will continue iterative upgrades, incorporating more unstructured data sources (such as earnings call audio, news sentiment) and exploring cutting-edge technologies like MCP (Model Context Protocol), Skill invocation, and prompt engineering to enable more complex automated analysis and task collaboration.
"AI Qing Fu" is not only the core engine of Ping An Fund's intelligent transformation but also aims to construct a next-generation investment research paradigm characterized by deep human-machine integration. It embarks on a new journey where artificial intelligence participates in value judgment and strategic decision-making, injecting powerful momentum into the technological empowerment and high-quality development of the asset management industry.