Motivation

The motivation behind ensuring financial security and having access to current information lies in the importance of making well-informed decisions and staying ahead in a rapidly changing environment. For professionals in the financial and insurance sectors, access to up-to-date data is crucial for effective risk management, regulatory compliance, and maintaining a competitive edge.

The introduction of a Web Search Assistant, empowers users to obtain real-time information through a chat interface, enhancing decision-making and problem-solving by providing the latest insights and data from the web. This dynamic access to live information helps professionals stay informed about the latest developments, ensuring they can navigate the complexities of the financial world with confidence and agility.

Goal

The goal of the Web Search Assistant is to provide real-time access to the vast and ever-changing information available on the internet through a chat interface. By seamlessly integrating live web searches, users can obtain accurate and up-to-date answers to their queries, enhancing their ability to stay informed, make informed decisions, and solve problems with the latest available data.

Structure and Logic of Assistant

1. Document Search Module:

   • The assistant includes a web search module, which is essential for accessing real-time data from the internet.

   • LLM Knowledge vs. Web Search: When users ask a question, the LLM first determines whether it can provide an accurate answer using its built-in knowledge. If the internal knowledge suffices, the assistant will respond without performing an online search, and no external references will be provided. However, if the question requires up-to-date information, the model will generate a search query based on the discussion and optimized for search engines, triggering a web search.

   • The LLM responds with the most relevant answer by retrieving information from the top search results of the search engine.

   • Relevant websites are appended to the answer as clickable references when a web search is performed.

2. Supported Search Engines:

   • Theoretically, the web search module can support any search engine as long as it offers API. For now, Bing and Google engines are both supported. You just need to provide the correct API keys.

3. Additional functionalities:

   1. Both Bing and Google support scoping the search domains. This means that the user can configure the list of trusted sources (domains) that can be returned by the search engine.

Possible Adaption of Assistant

The web search module can be customized in several ways:

• Enhanced Prompting: The instruction to generate the search query can be modified (e.g. to ask the model to always perform the search in English, which usually gives the best answers).

• Search Engine: Select a search engine (assuming it’s configured).

• Many other parameters can be changed. Please refer to the config documentation Web Search Module.

Required and Optional Modules

The following modules are required/optional for this assistant:

Example AI Assistant Configuration:

Download the provided TXT file and upload it into a new space as an AI Assistant configuration. This will create a web search assistant using GPT-4o:

Open

Prompt Engineering Guide

This assistant is set up such as it automatically generates an optimised search query based on the entire the discussion history. However, by following these guidelines, you can enhance the effectiveness of your prompts and queries, leading to more accurate and useful responses from the Web Search Module.

Triggering Online Searches

• The assistant can respond using its built-in knowledge or perform an online search.

• Default Response: Asking a question like “What is the New York Stock Exchange?” will be answered using the LLM’s knowledge.

• Forcing a Search: To trigger an online search, explicitly request it, e.g., “Please search online for the New York Stock Exchange,” and the assistant will retrieve the latest information.

Avoiding Leading Questions

Don’t suggest an answer within your query.

• Leading: “Is New York the best city for startups?”

• Open-ended: “What are the top cities for startups and why?”

By following these guidelines, you can ensure more accurate and effective responses from the Web Search Assistant.