Bedrock
Conversational
Agent
Use Case: Custom Conversational Agents
The Problem
The request we get most from clients is creating a conversational agent/chatbot based on their own data. The data sources can range from a single PDF file to a data warehouse of documents, spreadsheets, and SQL databases. These documents cannot simply be put into ChatGPT for several reasons:
-
ChatGPT does not provide enough accuracy when analyzing several documents at a time.
-
The data/documents are too large for ChatGPT.
-
The data is sensitive/proprietary and the chatbot needs to be hosted on internal infrastructure.
-
The chatbot needs to use current information/needs to be continuously updated
-
The outputs of the chatbot need to be fed into another software.
-
Other LLMs may perform better than ChatGPT.
-
The chatbot needs to be hosted on an external website
The Solution
The solution to these issues is creating customized conversational agents trained on private data using Retrieval-Augmented Generation.
​
Retrieval-Augmented Generation (RAG) agents combine the ability to generate natural language responses with accurate information retrieval from large document collections. They analyze your question, search for relevant data, and generate precise, up-to-date answers. This approach ensures accurate, current, and specialized responses without compromising data security and privacy.
​
These agents can be hosted on internal infrastructure, preventing sensitive information being sent externally via API. The outputs of these chatbots can be fed into other internal/external software, handle large amounts of data of various types, and provide much better results than ChatGPT because of fine-tuning.
​
By bringing all of this together, RAG-powered conversational agents let your team work faster, reducing time spent searching through documents, and allowing you to make better decisions with confidence. They turn large collections of internal files into an easy-to-use conversational tool that delivers answers instantly, giving your organization a secure, efficient, and highly scalable way to unlock the value of your private data.​
The Diagram
The architecture diagram below shows an example of how we build intelligent conversational agents using AWS Bedrock. The documents are stored securely in Amazon S3, and the system reads and learns from them so it can give accurate, personalized answers.
​
When you ask the agent a question, it goes through a secure gateway and into the conversational engine. Behind the scenes, the system uses Amazon Bedrock to understand the question, look up the most relevant information from your documents, and craft a clear response. It does this by searching a special index, powered by Amazon OpenSearch, that helps the chatbot quickly find the exact details it needs.
​
To keep responses safe, consistent, and trustworthy, we are also able to guardrails and a custom set of instructions that guide how the chatbot should behave.
​
The result is a smooth, secure conversational experience where you get fast, reliable answers comparable to using ChatGPT!
The Feedback
“Vyrtices was able to securely ingest all our data to make a fantastic internal conversational agent. We were all very impressed."