3.4 Chat and Search Capabilities

Orbit Insight is equipped with advanced chat and search capabilities, allowing users to efficiently find and analyze relevant information across vast datasets. This chapter explores how these capabilities work, from the underlying technology to the advanced features that enhance the user experience.

1. What Does Orbit Search Over?

Orbit Insight searches across all the knowledge bases provided by the platform, as well as the internal knowledge bases of clients. This comprehensive search capability ensures that users can access both external data and their proprietary information in a single, unified platform, making it easier to conduct thorough and informed research.

2. How Does the Search Work?

The search functionality within Orbit Insight is powered by a sophisticated process that involves several key steps:

• PDF Pre-Processing: The search process begins with PDF pre-processing, where clean text is extracted from documents. This step ensures that all textual information is accurately captured, regardless of the original format.

• Embedding Models and Vectors: After text extraction, embedding models are applied to generate vectors for the text. These vectors represent the semantic meaning of the text in a numerical format.

• Indexing: Both the clean text and the vectors are indexed into the search engine. This dual indexing ensures that the system can efficiently handle both keyword-based searches and more complex semantic queries.

• Search Execution: When a user enters search terms—whether keywords or full sentences—those terms are also converted into vectors. The search engine then compares these vectors with the indexed vectors of the text blocks, identifying the most relevant matches based on their similarity.

3. How Are the Results and Relevance Determined?

The relevance of search results in Orbit Insight is primarily determined by the cosine similarity between the vector representation of the search terms and the vectorized text blocks within the knowledge bases.

• Cosine Similarity: This metric measures the cosine of the angle between two vectors, with smaller angles indicating higher relevance. The closer the vectors are in direction, the more semantically similar they are, resulting in a higher relevance score for the associated documents.

• Relevance Scoring: The search engine uses these relevance scores to rank the search results, ensuring that the most relevant blocks of text and their corresponding documents appear at the top of the results list.

4. How Copilot Chat Happens

Orbit Insight’s copilot chat functionality leverages the power of LLMs (Large Language Models) to provide users with accurate and contextually relevant answers to their questions.

• Initial Search: When a user asks a question, the system first conducts a search within the search engine to identify relevant blocks of text along with their associated metadata.

• LLM Processing: The system then sends both the user’s question and the relevant text blocks to the LLM. The model processes this information to generate a coherent and informative response, which is then returned to the user.

• Integrated Response: This integration of search and LLM processing ensures that the chat responses are not only accurate but also grounded in the specific documents and data relevant to the user’s query.

5. Advanced Features

Orbit Insight’s chat and search functionalities are enhanced by several advanced features designed to improve the relevance and precision of search results.

• Question Intention Detection: The system includes an algorithm that detects the intention behind a user’s question. This feature enriches the query by understanding its context and aligning it more closely with investment research topics, ensuring that the results are more relevant to the user’s needs.

• Enhanced Query Understanding: By refining the user's query, the system can provide more accurate and contextually appropriate search results, improving the overall search experience.

6. Why Vector Search?

Vector search is a crucial feature of Orbit Insight, particularly for handling multilingual documents. By converting all text into vectors, the platform can perform true multilingual searches, where documents in different languages can be searched and compared on a semantic level.

• Multilingual Capability: This approach ensures that search results are consistent across languages, maximizing the value of Orbit Insight’s global document coverage. Users can search in one language and find relevant documents in multiple languages, breaking down language barriers in investment research.

• Semantic Consistency: Vector search allows the system to focus on the meaning of the text rather than just the exact words, making it possible to find relevant information even when the query and documents are in different languages.