Dept. of Veterans Affairs Chatbot

VA Chatbot is a self-service tool on VA.gov that helps veterans problem-solve and complete tasks with little to no human intervention, scaled to approximately 10K users per month.

Approach

Led a large cross-functional team to build and iterate on the chatbot in close collaboration with VA stakeholders, integrating Microsoft’s conversational AI stack with custom NLU and knowledge retrieval capabilities to handle an expanding set of veteran use cases.

Highlights

Role: Technical Lead Led delivery with a team of 14 (10 Devs, 1 QA, 1 XD, 1 ProjM, 1 ProdM).

• Scaled the chatbot from initial launch to ~10K monthly users, contributing to measurable call center volume reduction.
• Integrated Azure Cognitive Services (Custom Question Answering, Conversational Language Understanding) and OpenAI APIs to expand the chatbot’s ability to handle complex veteran queries.
• Built the frontend experience with React/Redux and Microsoft Webchat, backed by a Ruby on Rails API.
• Managed infrastructure provisioning and deployment with Terraform and Docker across the VA’s cloud environment.
• Collaborated with Data Scientists and ML Engineers to refine the chatbot’s Natural Language Understanding model using ~200K conversation transcripts, improving the accuracy of intent recognition and response quality.

Tech Stack

Microsoft Copilot (formerly Power Virtual Agent), Microsoft Bot Framework SDK, Microsoft Power Platform, Azure Cognitive Services, OpenAI APIs, React/Redux, Ruby on Rails, Microsoft Webchat, Jest, Terraform, Docker

Outcomes

The chatbot is live in beta on the VA.gov contact us page, with steadily increasing adoption and a growing feature set. Directly contributed to reducing call center volume by giving veterans a reliable self-service channel.