Gloucester County, Virginia serves roughly 40,000 residents across a range of county services, including land assessment, building permits, and code compliance. When residents needed permit guidance, the process often sent them through three or four departments before they got a complete answer. This delayed answers for residents and created an administrative burden for county staff.

The county had explored AI once before, but that effort stalled when proposed costs exceeded county budgets. Gloucester needed a path that fit within a realistic government budget and could still deliver a production-ready system.

Tech 42 built an AgentCore-based AI agent on AWS that lets residents query parcel records and permit requirements in plain language. The first phase went live on budget, at roughly 20% the cost of prior proposals.

Results:

• ~85% less cost to run in production compared to other vendors
• Agent serves 40,000 Virginia residents and 3,600 businesses with 24/7 access to permit and parcel guidance
• ~1,500 estimated monthly citizen interactions at full public release
• Serverless architecture automatically scales up or down, delivering fast answers and cost efficiency

“Since rolling the solution out to internal subject matter experts, Gloucester County has seen it become a genuine time saver and efficiency gain for staff who previously had to manually search through the same materials to answer resident questions. It's also helped ease reliance on a small number of long-tenured staff who hold critical institutional knowledge, making that expertise more accessible across the team, while surfacing new opportunities to close integration gaps between departments.”
- Britton Hogge, Senior Software Analyst

Challenge: Multi-department friction and a stalled previous attempt

Gloucester County staff described a familiar government service problem. Residents coming in for a building permit often had to visit multiple departments before getting everything they needed. Sheds, in particular, created consistent confusion because the requirements vary based on lot size, setback rules, and zoning classifications. And the data needed to answer questions is fragmented and lives across multiple departments.

Technical constraints added more challenges. The county's parcel assessment database ran on an on-premises server that was already operating near capacity. And strict data controls meant view-only access to the database, with no exposure of confidential records. Any AI integration would need to work within those limits.

The county had tried to address this before. But with prior proposals exceeding budget amounts, County leadership declined, and the project sat dormant. Tech 42 provided a complete solution that was within the county’s budget, leveraged AWS funding assistance to offset costs, and worked within the county's procurement process.

Solution: An AI agent built for government constraints

Tech 42 designed an auto-scaling, two-tool AI agent using Strands + AgentCore, that’s structured to work within the county's data governance and procurement requirements.

The first stage of the agent is a text-to-SQL tool that converts resident questions into SQL queries against the county's parcel database. The agent returns answers in plain language. A resident can ask "what is the assessed value of 123 Main Street" without knowing how the database is structured or how to write technical queries.

A second tool handles retrieval-augmented generation (RAG) against county documentation stored in Amazon S3. Permit requirements, ordinance references, and department procedures live in a Bedrock Knowledge Base backed by S3 vectors. When a resident asks about building a shed, the agent retrieves the relevant requirements rather than generating an answer from model knowledge.

Both tools run through AWS AgentCore Runtime. Langfuse handles monitoring and tracing. Bedrock Guardrails filter sensitive data and content before any response leaves the system.

And what about the on-premise database constraint? Tech 42 ensures the assessor's data stays on-prem and read-only. The agent only queries a controlled snapshot.

In the end, the county received a fully deployable solution that meets AWS best practices. But just as important, the engagement ended with full documentation and knowledge transfer so the county staff is confident in using the agent moving forward. 

Architecture overview

Figure 1: Gloucester County AI agent high-level design (HLD).

Component summary

‍

Outcome: 24/7 citizen access at a fraction of prior proposals

Gloucester County launched the internal validation phase against real-world queries and edge cases with a small group of county staff. From there, access will expand to the public.

For residents, the change means getting permit guidance at any hour without navigating department phone trees or waiting for office hours. For county staff, the agent handles the repeat, high-volume questions that currently require manual time to answer.

“Working with Tech 42 has felt less like managing a vendor and more like having a true technology partner — one that delivered exactly to scope, communicated clearly and promptly throughout, and brought genuine expertise to every part of the engagement.”
- Britton Hogge, Senior Software Analyst

Teams evaluating similar needs in the public sector can review Tech 42's AI agent services or use the AgentCore Cost Calculator to estimate infrastructure costs before committing to a scope.

About Gloucester County

Gloucester County is a county government in Virginia, serving a population of approximately 40,000 residents and 3,600 businesses. County services include land assessment, building and zoning, and permit processing. Gloucester has pursued technology modernization with a focus on responsible procurement and constituent service quality.