A large U.S.-based brokerage and wealth management client has a large number of users for its retail trading platform that offers sophisticated trading capabilities. Although extensive information was documented in hundreds of pages of product and process manuals, it was difficult for users to access and understand information related to their specific needs (e.g., How is margin calculated? or What are Rolling Strategies? or Explain Beta Weighting).

Our Gen AI solution encompassed:

• Building a user-friendly interactive chatbot using Dialogflow in Google Cloud
• Ringfencing a knowledge corpus comprising specific documents to be searched against and summarized (e.g., 200-page product manual, website FAQ content)
• Using a vector database to store vectors from the corpus and extract relevant context for user queries
• Interfacing the vector database with OpenAI API to analyze vector-matched contexts and generate summarized responses

The OpenAI GPT-3.5 turbo LLM (170 bn parameters) delivered impressive linguistic search and summarization capabilities in dealing with information requests. Prompt engineering and training are crucial to secure those outcomes.

In the case of a rich domain such as a trading platform, users may expect additional capabilities, such as:

• API integration, to support requests requiring retrieval of account/user specific information, and
• Augmentation of linguistic approaches with semantics to deliver enhanced capabilities.