Adding More Data to Your GenAI Service
Posted: Thu Feb 13, 2025 3:27 am
Combining these approaches – fine tuning and RAG – can provide better quality responses to users. Using RAG is also beneficial when you have strong data privacy requirements to meet and you do not want to store your company IP or customer PII in your LLM models.
Injecting data via RAG is the most efficient way to brazil whatsapp number data provide context information to your generative AI system that was not present in the model’s training data set. This is especially useful in cases when you have data that updates regularly and your users will want more recent data in their responses. This data for fine-tuning and RAG comes from your existing data sets, such as your databases, customer relationship management, enterprise resource planning, and knowledge management systems. However, it can also come from less structured sources like mail conversations, speech recordings of service calls, videos, images, and more.
To manage this data, AI agents will require a storage layer for their short-term and long-term memory. As AI agents are stateless, the short-term memory keeps a record of a conversation and uses that data to generate further responses. This acts like a memory stream with a large number of observations that are relevant to the agent’s current situation, including a log of previous questions and responses. One approach to efficiently support this is to use vector search to support retrieval. This also allows you to manage the history that is used as memory for the agent, have full control over the data lifecycle, and define any permissions or security rules.
Injecting data via RAG is the most efficient way to brazil whatsapp number data provide context information to your generative AI system that was not present in the model’s training data set. This is especially useful in cases when you have data that updates regularly and your users will want more recent data in their responses. This data for fine-tuning and RAG comes from your existing data sets, such as your databases, customer relationship management, enterprise resource planning, and knowledge management systems. However, it can also come from less structured sources like mail conversations, speech recordings of service calls, videos, images, and more.
To manage this data, AI agents will require a storage layer for their short-term and long-term memory. As AI agents are stateless, the short-term memory keeps a record of a conversation and uses that data to generate further responses. This acts like a memory stream with a large number of observations that are relevant to the agent’s current situation, including a log of previous questions and responses. One approach to efficiently support this is to use vector search to support retrieval. This also allows you to manage the history that is used as memory for the agent, have full control over the data lifecycle, and define any permissions or security rules.