Portfolio managers in the asset management industry face challenges in identifying secondary and tertiary impacts on their portfolio companies originating at suppliers, customers, partners, or other entities in their ecosystem. This piece illustrates an automated system that combines knowledge graphs and generative AI to help in identifying such risks. The system scans real-time news and cross-references it with relationship maps.
The system firstly constructs knowledge graphs building intricate relationships between companies. It then uses these graphs combined with generative AI to detect secondary and third-level impacts from news events. This can, for instance, highlight the potential disruption to an auto manufacturer’s production line due to delays at a parts supplier, even if not directly referenced.
The solution can be deployed on AWS, which offers servers less, scalable, event-driven architecture. It uses two key AWS services, Amazon Neptune, and Amazon Bedrock. Neptune is a graph database service to help build and run applications that work with highly connected datasets. Amazon Bedrock is a managed service that offers a choice of high-performing foundation models from leading AI companies.
In addition to these resources on AWS, the solution also discusses using a company’s annual reports to build a knowledge graph. This process can be automated using generative AI services like Amazon Bedrock. The structured data from these unstructured documents is extracted using large language models (LLMs). Finally, disambiguation is conducted to clarify the extracted information.
The next step in the solution involves processing news articles and enriching portfolio managers’ news feeds. Managers can subscribe to any news provider. An ingestion pipeline processes articles pulling entities, attributes, and relationships. Amazon Bedrock extracts these elements and disambiguates against the knowledge graph to identify the corresponding entity. Any connections between the content of the articles and portfolio companies are highlighted, with the final output being an enriched news feed.
The architecture of the solution workflow includes steps such as uploading official reports, the extraction of key information using Amazon Bedrock, the establishment of connections to the knowledge graph, and the selection of links to monitor.
This solution provides a promising direction for AI and graph databases in weaving together different sources of data to produce actionable insights. This may be particularly beneficial for investment professionals needing to quickly ascertain potential risks in their portfolios. The posts offer information on deploying this prototype. Although the model is in an early stage of development, it has the potential development in a direction that would significantly streamline the work of portfolio managers.
