Active equities manager was looking to streamline its memo drafting with an interactive investment knowledge co-pilot to assist analysts in extracting and synthesizing insights across diverse documents.
We mapped the investment process to define requirements for UI, workflow, response time and data management.
We built capability to extract knowledge from diverse documents, tables, public sources such as Google and leveraged LLMs such as GPT-4.
We worked with the investment teams to ensure the co-pilot created differentiated insights from multiple sources and value beyond retrieving information.
Critically, we worked to demonstrate the accuracy of the outputs and the steps to get there., ensuring explainability of the plan, intermediate steps and references.
We could then enhance the reliability and performance of the solution by limiting the search area.
We delivered a set of LLM agnostic capabilities that enhanced the performance of current off-the-shelf LLM tools — allowing us to embed investment process knowledge through the planning and execution agents.
We deployed the preliminary suite of capabilities for an end-to-end Investment co-pilot workflow with the following capabilities:
- Conversational interface for question answering across multiple files and company clusters
- Summarization and key point extraction from multi-modal data sources and documentation
- Proprietary planner agent to support interrogation and questioning from analysts (e.g., compare and contrast)
- Tool integration to retrieve information from web and access 3rd party tools (e.g., Wolfram Alpha)
In addition, we up-skilled the internal data science team on technical best practices through day-to-day collaboration and on-the-job coaching.