Energy Pricing Assistant

• Energy Pricing Assistant
  • Overview
    • 1. Bedrock Flows
    • 2. RAG Knowledge Base
    • 3. Runtime Chatbot
  • End-to-end Architecture
  • Demo Script
  • Setup
    • Pre-requisites
    • Run CDK
    • Trigger the data simulation
    • Cleanup

Overview

This demonstration showcases an AI assistant designed for gas station pricing analysts. The assistant can gather and analyze data from diverse sources, such as weather patterns, traffic conditions, historical pricing strategies, and competitor actions. By leveraging this information, analysts can significantly accelerate the process of estimating pricing updates and uncover valuable insights that may inform decision-making.

1. Bedrock Flows

Near realtime data sources are queried on demand through Bedrock Flows. Bedrock Flows enables analysts to ask for fuel price estimates using several data sources from dynamodb or potentially other sources or vendor APIs like weather and traffic. With context gathered through this flow, a prompt is constructed for Amazon Nova Lite v1 via Bedrock to provide a fuel price estimate and supporting evidence. Anthropic Claude Haiku v3 is also utilized in generating a response to results retrieved from the Bedrock Knowledge Base.

2. RAG Knowledge Base

The RAG knowledge base leverages the Knowledge Bases for Amazon Bedrock and its metadata filter feature, allowing for more granular search on pricing data based on metadata like location and other relevant attributes. This knowledge base stores and organizes the ingested strategy data in a way that supports efficient retrieval and generation of information based on the RAG model, while enabling users to slice and dice the data according to their specific needs.

3. Runtime Chatbot

The runtime chatbot is a React-based website that uses a WebSocket API and a Lambda function architecture. The conversation history is stored in DynamoDB. The Lambda function uses the Amazon Bedrock Converse API to reason and retrieve relevant documents from the knowledge base, providing the final answer to users in training scenarios for fuel pricing analysts.

End-to-end Architecture

Demo Script

The demo script, use case, and persona is provided here: Demo Script

Setup

Pre-requisites

1. The solution is only available in us-east-1. Please choose either of them to proceed.
2. Enable models in Amazon Bedrock: for this use case, you need to enable Amazon Nova Lite v1, Anthropic Claude Haiku v3, Titan Text G1 - Premier, and Titan Embeddings G1 - Text
3. Get started with CDK.
4. Install Docker. Because we are bundling Lambda functions when running CDK so we need to install Docker. Please see the blog post about Building, bundling and deploying applications with the AWS CDK

Run CDK

1. Clone this repository.
2. Export AWS_DEFAULT_REGION to us-east-1 region following the CDK cli guide.
   hljs language-ini

   Copy

   export AWS_DEFAULT_REGION=us-east-1
   

3. Install npm modules
   hljs language-bash

   Copy

   cd cdk-stacks
   npm install
   

4. CDK bootstrap
   hljs

   Copy

   cdk bootstrap
   

5. At this point you can now synthesize the CloudFormation template for this code.
   hljs language-arduino

   Copy

   aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws
   cdk synth
   

6. Deploy the application
   hljs language-css

   Copy

   cdk deploy --all --require-approval never --parameters EnergyPricingAssistantAgentStack
   

Trigger the data simulation

1. Sync Bedrock Knowledgebase

   a. Open AWS Bedrock console. Knowledge base -> EnergyPricingAssistant.

   b. Click sync

2. Trigger the data simulation for initial use:

   a. Open AWS Lambda console. Find the Lambda function that looks like this EnergyPricingAssistantAge-StationDataGenerator27AA-KtrEwaUxCFzT

   b. Click on Test to trigger the data generation. This takes 3-5 minutes. You will need to do this twice

Cleanup

Run the following commands to destroy all Stacks.

cdk destroy --all

Enter y upon the prompt to destroy each Stack.

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.