GenAI Conversational Agent

## Challenge

EfectoLED integrates in their websites Zendesk as chat application, with this they can offer detailed assessment of the products to the end-users.

The main challenge was produced because of the need of chat automation since it was a real person who was in charge of answering the doubts customers had. Other key points to be considered was the formality answering and detection of sensitive data.

## Solution

The Magic Beans work around solution was to implement a GenAI chatbot using Amazon Bedrock. Bedrock contains several generative models like LLMs, for the solution Anthropic Claude v2 was implemented due to its inherent resistance against sensitive data and conversational tone.

With this, the speed of answering questions was improved from minutes to just very few seconds.

## Results and Benefits

**Cloud Deployment:** Elasticity can be achieved through an implementation of a serverless event-based architecture with AWS Lambda.

**Cost Optimization:** As AWS infrastructure was implemented using serverless services there is no need to maintain and update them, also the pay-as-you-go feature allows to pay for only what is consumed.

**Agility:** With Amazon Bedrock, a solution that could be addressed in several months, was speed up and running in few days thanks to the use of Foundational Models.

**Visibility:** Amazon offers several visibility real-time metrics in order to keep trace of usage with custom dashboards.