DCA demos chatbot proof‑of‑concept using AWS Bedrock; CTO outlines GenAI policy, human review and 90‑day pilot timeline

The Department of Consumer Affairs’ Office of Information Services presented a proof‑of‑concept chatbot for the California Veterinary Medical Board website and described how state GenAI policy will shape procurement, deployment and oversight.

Bayard Cowan, chief technology officer at DCA, and Keith (senior web architect) demonstrated a chatbot built using Amazon Web Services and Bedrock during the board meeting. Cowan described the wider procurement and policy context for GenAI in state government, citing an executive order he identified as "n 1223," a California Department of Technology technology letter referenced as "25‑01," references to Senate Bill 896 and Assembly Bill 2013, and SAM guidance the presenters described as section 4986. Cowan said DCA has completed updated acceptable‑use and GenAI training policies and reported that roughly 72% of DCA staff had completed GenAI training to date.

Cowan said the POC is deliberately constrained to public website data and configured to avoid indexing enforcement or other sensitive documents; the model will only return answers rooted in the selected website content and will surface links to pages for users to review. He emphasized the requirement for human oversight: chat histories are logged and back‑office staff must review interactions and tune responses so the chatbot’s answers match agency policy and legal requirements. "The biggest thing that I've seen with the state of California and the GenAI is being very careful as to make ensuring that it's not grabbing data out on the Internet," Cowan said.

Keith demonstrated the chatbot’s features: automatic generation of candidate question‑and‑answer pairs from selected webpages, the ability to map multiple ways a user might ask the same question (utterances) to one canonical answer, in‑chat links that open in new windows, thumbs‑up/thumbs‑down feedback, and analytics that surface frequently asked or unanswered questions for staff to address. The system matched informal phrasing and abbreviations (for example, "vet" for "veterinarian" and acronyms such as "RVT") and supported branching prompts that can ask clarifying questions and guide a user to the right procedure or page depending on answers.

Board members asked about staffing and maintenance; Jessica (board staff) said the board recently approved a reclassification for a data‑analyst position intended to support reports and would be the primary person responsible for monitoring and maintaining the chatbot content. Cowan said the team expects to be able to begin development and a staging deployment within roughly 90 days but that DCA is also evaluating Google Cloud, Microsoft and the state's planned statewide digital assistant before final decisions on production use.

The board discussed outreach plans to direct callers to the chatbot and asked whether the tool would reduce phone and email workload; staff said it should reduce routine inquiries once the chatbot is built and promoted and that the analytics will help identify high‑volume issues for staff to address. The board opened the item for public comment; no remote participants requested to speak.

DCA said next steps include further evaluation of alternative cloud providers and the statewide digital assistant, a formal GenAI risk assessment and procurement steps required by DOT and DGS before broader production rollout.