Skip to main content
Back to Repository

AI Land Due-Diligence SaaS

Applied AI Platform
Role: Applied AI Engineer (Contract)Year: 2026-Present

Client project: Confidential Client (Contract)

Technical Deep Dive

Architecting AI-agent workflows and the AWS platform for an established business launching an AI-integrated SaaS product that scores land parcels for purchase risk and assembles due-diligence packages, partnering directly with the client's tech lead and founder.

Client Context

The client arrived with a draft infrastructure spec, an open AWS region choice, and a tech lead who wanted AI coding agents doing much of the work, with human gates. The architecture had to be settled on evidence, fast, and agents had to do real work without being able to merge, release, or touch production on their own.

Execution

Architecting a five-milestone Terraform platform (landing zone, networking, Aurora PostgreSQL with PostGIS, ECS Fargate CI/CD, Bedrock Claude). Built a PR gate that blocks merges until the Linear ticket clears human spec approval, 6 Claude Code hooks that stop agents on failing checks or default-branch edits, and 4-hour MFA-backed IAM role sessions for agents, inside a harness of role-separated agents (driver, adversarial reviewer, release verifier, doc checker) and a cross-model review MCP server.

Outcome

Led the us-west-2 region decision with the tech lead: 4 Fargate AZs, full Bedrock AgentCore support, and $235/month cheaper. The research run verified 44 of 45 spot-checked claims in 65 minutes, and the harness runs 290+ passing test assertions. Platform milestones are in progress.

Why It Mattered

The adversary paid for itself. Agents are fast at producing claims and bad at doubting them, so the adversarial reviewer and the human gate are the parts worth never cutting: speed only helps when something is paid to ask what we got wrong.

“Architecting a five-milestone AWS platform in Terraform and a guarded AI-agent delivery workflow where agents open PRs and humans approve specs and merges; a 15-agent research run verified 44 of 45 spot-checked claims in 65 minutes.”

Core Stack

AWS
Terraform
Amazon Bedrock (Claude)
ECS Fargate
Aurora PostgreSQL + PostGIS
AI Agent Orchestration
MCP
Human-in-the-Loop Gates

Metrics

terraform_milestones

5

claims_verified

44 of 45

research_run

65 min

test_assertions

290+

region_savings

$235/month

Evidence

Client work under confidentiality. Happy to walk through it in a conversation.