Applications of GenAI

Within the Talkative platform, GenAI is used in two primary areas:

1. Knowledgebase-Driven Use Cases:
   Examples include chatbot responses and copilot functionalities—here, you can choose your AI model.

2. General Use Cases:
   Examples include interaction summaries—currently, these do not offer model choice. Regional-specific open-weights models are used here (AWS-hosted LLAMA).

Below is a summary of the GenAI features and whether you can choose your model:

GenAI Architecture

1. Knowledgebase-Driven Use Cases

Setup

• Upload Knowledgebase:
  Upload your knowledgebase source material to Talkative.

• Storage:
  The source material is stored as a text file within Talkative’s regional AWS S3 bucket.

• Large Datasets:
  For larger datasets, AWS Cohere is used to generate a vector store of your knowledgebase data.

• Model Selection:
  Choose your preferred LLM for runtime.

• Configuration:
  Enable the feature by configuring the appropriate settings (e.g., Chatbot or Copilot configuration).

Runtime

• Prompt Generation:
  The customer’s message, conversation transcript, and relevant knowledgebase data are combined into a prompt sent to the selected LLM.

• Embedding Model:
  AWS Cohere is utilized as the embedding model.

• Response Integration:
  The response from the LLM is integrated into the system’s reply (whether for direct customer interaction or within a copilot message).

2. General Use Cases

Setup

• Feature Enablement:
  Activate the feature within Talkative’s settings.

• Model Limitation:
  Currently, there is no option to choose your LLM for these features. All general use cases are powered by AWS LLAMA.

Runtime

• Processing:
  Depending on the feature, the text is sent to one of the LLMs (AWS LLAMA, AWS Anthropic, OpenAI, or Google).

• Response Integration:
  The LLM’s response is used to generate interaction summaries, rephrase messages, or perform other functions.

Data Processing and Privacy

• AWS Bedrock (LLAMA, Anthropic Models):
  AWS does not share your data with model providers. Additionally, Bedrock does not store or log your prompts or completions.
  Learn more: AWS Bedrock Data Protection

• OpenAI:
  OpenAI does not use your data to train its models. Data is stored for 30 days. Please contact us if you require a different data retention period.
  Learn more: OpenAI API Data Usage Policies