-
-
ComplianceScout Thumbnail — Autonomous Professional Compliance Agent.
-
A contractor inputs a plain language project description, and immediately initiates tool execution to review local San Francisco building
-
ComplianceScout drafts structured markdown permit applications but strictly enforces human review, never submitting files automatically.
-
ComplianceScout drafts structured markdown permit applications but strictly enforces human review, never submitting files automatically.
-
Powered by Strands SDK & Amazon Bedrock AgentCore. Features persistent session memory & OpenTelemetry tracing.
Inspiration
Independent contractors—the people actually building our homes and upgrading our infrastructure—lose over 10 hours every single week buried in administrative paperwork. Before they can even pick up a tool, they have to navigate dense, fragmented municipal zoning laws, cross-reference complex building codes, and fill out meticulous permit application forms. We built ComplianceScout to clear this bureaucratic runway, allowing skilled experts to focus on physical construction rather than tedious paperwork.
What it does
ComplianceScout is an autonomous Professional Agent designed for contractors, electricians, and independent tradespeople. Give it a plain-language project description (e.g., building a 12x14 deck or upgrading an electrical panel), and it automatically:
- Searches local municipal zoning and building code databases via custom tools.
- Analyzes structural requirements (like setbacks, guardrails, and C-10 licensing).
- Drafts structured markdown permit application packages.
- Implements strict guardrails: it never submits anything automatically, setting a mandatory
requires_signature: trueflag and prompting the human professional for review and sign-off.
How we built it
We engineered ComplianceScout using modern agentic infrastructure:
- Strands Agents SDK: Built the core agent harness loop using a custom system prompt,
SlidingWindowConversationManagerfor bounded conversation context, andSequentialToolExecutorfor deterministic, auditable tool sequencing. - Amazon Bedrock & Claude Sonnet: Leveraged state-of-the-art foundation models via Bedrock to power natural language reasoning and code parsing.
- Amazon Bedrock AgentCore: Deployed serverlessly with AgentCore Runtime, utilizing AgentCore Memory for persistent cross-turn contractor conversations and AgentCore Observability for OpenTelemetry tracing.
- Streamlit UI: Created a responsive, contractor-friendly web interface for real-time interaction.
Challenges we ran into
- API and Model Throttling: Managing sandbox daily token limits and transitioning smoothly between regional inference profiles and direct regional Bedrock endpoints.
- Deterministic Tool Ordering: Ensuring the agent didn't interleave multi-tool calls in a confusing way, which we resolved by enforcing strict sequential execution hooks via the Strands SDK.
Accomplishments that we're proud of
- Successfully built an end-to-end agent that moves past simple chat into true structural workflow automation.
- Deep integration of AgentCore Memory ensuring the agent remembers missing variables (like deck height or panel amperage) across separate conversation turns.
- Designing a robust human-in-the-loop guardrail that explicitly prevents accidental or unauthorized automated permit filing.
What we learned
We gained deep expertise in building production-ready agent loops using the Strands SDK, configuring OpenTelemetry tracing hooks for compliance auditing, and orchestrating serverless agent runtimes with Amazon Bedrock AgentCore.
What's next for ComplianceScout
- Integrating live municipal GIS and open-data zoning APIs across major metropolitan areas.
- Expanding support for multi-trade permit bundling (mechanical, plumbing, and electrical combined).
- Enhancing team collaboration dashboards for general contractors managing multiple sub-contractors simultaneously.
Log in or sign up for Devpost to join the conversation.