Full Stack Developer
American Automobile Association
Insurance company that provides car insurance and roadside assistance.
Team 2
GenAI Chatbot (AAA Innovation Award Winner 🏆)
Date
06/2022 - 06/2024
Team 1
Insurance Self Service
Date
06/2024- Current
Insurance Self Service Team
My team delivers web apps that enable over 16 million automobile insurance holders across the United States make changes to their policies. They bring in over $3 billion annually. I contributed to two of these applications.
I wrote the front-end in typescript, with redux for state management. I wrote the API specs in Swagger.
I wrote the page where users can select a vehicle to remove from their car insurance.
I integrated the drawer component that lets users navigate to different points of the flow. ADA compliant.
Library: Material UI
I wrote lambdas which perform business logic by communicating with internal services. I exposed them to the app via API Gateway.
Infrastructure
Deploy code to development and production environments following continuous integration / continuous delivery principles.
Write infrastructure as code (IaC), such as code pipelines and cloudformation stacks, to deploy lambdas, web builds and gateways.
Language: Typescript
Migrate IaC from AWS to Github workflows to move our repositories from Github Teams to Enterprise.
Language: YAML
Release apps to different markets using feature flags in split.io.
Generative AI
I worked on a generative AI Chatbot that Emergency Roadside Service agents can use to inquire about their manual. It leveraged a large-language-model (LLM) with retrieval-augmented-generation to output an answer.
Retrieval-augmented-generation is a prompt engineering technique that takes a user’s input and compares it against documents in a knowledge base. The most similar ones get injected into the prompt for the LLM to summarize. I wrote the logic for this using the cosine-similarity algorithm.
I contributed to the lambda that made an API call to OpenAI’s Large Language Model using the LangChain library. It used MongoDB to store the vectors, a representation of a sentence into numbers.
Language: Python
I wrote the schema for the logs that recorded the agent’s query, response from the LLM, and the agent’s feedback about it. The team used this to improve the LLM’s accuracy to a production-ready level.