Walter is an artificially intelligent bot that creates and sends customized newsletters to subscribers at 7:00am sharp about the markets they're following. WalterAI gathers market data from various APIs for each user and their interested stocks and feeds the data into Bedrock to get AI insights from the LLM Meta Llama 3. This allows WalterAI to create tailored newsletters for each subscriber including information only about the markets relevant to the user's portfolio.
WalterAIBackend is the backend service that maintains the database of subscribers and their interested stocks as well as the service responsible for creating and sending the customized newsletters. WalterAIBackend is powered completely by AWS and runs on Lambda.
WalterAIBackend uses email templates stored in S3 to create the emails to send to subscribers. Each email template is written in HTML and uses Jinja to parameterize the inputs. WalterAIBackend pulls the email templates from S3 and renders the template given the templatespec.yml and the template.jinja file. The templatespec.yml is the specification file that tells Walter the name of the parameters as well as the Bedrock prompts to use to get the parameter value. An example of a templatespec.yml file is given below:
version: 0.0
#########################
# DEFAULT TEMPLATE SPEC #
#########################
TemplateSpec:
###########
# CONTEXT #
###########
Context:
User: {{ user }}
PortfolioValue: {{ portfolio_value }}
Stocks:
{% for stock in stocks %}
- Symbol: {{ stock.symbol }}
Price: {{ stock.price }}
{% endfor %}
########
# KEYS #
########
Keys:
- Key: User
Value: {{ user }}
- Key: PortfolioValue
Value: {{ portfolio_value }}
###########
# PROMPTS #
###########
Prompts:
- Key: Newsletter
Prompt: |
Write a newsletter with business casual humor!
MaxGenLength: 600After getting the answers to the prompts given in the templatespec.yml file, Walter renders the template with
Jinja and then sends the custom newsletter to the subscriber!