# How Kleo SDK Works

# User Owned Corpus

The corpus consists of multiple content owned by the user as captured from the LLM.

# Contextual Activity Classification Model

THis takes input from the User Owned Corpus captured creates a contextual classification based on most likely activity being performed. This in turn generates the activity chunks of specific activity.

# Activity Chunk Outputs

Multiple activity chunks are generated as outputs from the Contextual Activity Classification Model. Each activity chunks contains a lot of documents along top keywords in that activity.

# Reranking Algorithm for Pages

• Receives Activity Chunks as input.
• Now based on reranking algorithm using Term Frequency / Inverse Document Frequency and Gemini Vector Embeddings, the pages are reranked based on most relevance.
  • Embedding Model
  • TF-IDF
• Entities already extracted from the page corpus, for top 10 documents is then sent to LLM once again.

# LLM - Large Language Model

It takes input from, - Reranking Algorithm for Pages to get the most relevant pages with context. - SDK Prompt: A prompt from the SDK specifying additional requirements. The LLM processes the information and provides the most suitable user-owned data considering the given context and SDK prompt.

# Result

• Outputs the final result based on the decision made by the LLM.