I build AI agents that run inside real business operations - and make them safe enough for a compliance team to sign off on.What I deliver:- Claims and document intake: automated first-pass checks, extraction and routing on incoming claims or applications, with low-confidence files held for a named human reviewer instead of guessed at.- Accounts payable automation: three-way matching across purchase orders, receipts and invoices, with hard spend ceilings on any automated approval.- Customer support automation: answers drawn only from your approved policy corpus, with anything outside declared scope escalated to a person by name.- Workflow and system integration: connecting the tools you already run - CRM, ERP, ticketing, email intake, SFTP, SQL, object storage - over REST APIs and webhooks.How I work differently: every automated run passes five gates before it can touch your systems. Declared scope, a role manifest of what the agent may and may not do, hard limits and a spend ceiling checked before each metered call, human review on anything low-confidence, and an append-only sealed record of what happened. When a client, auditor or regulator asks what the automation did, you can show them, run by run.Tech stack: n8n, Make, Zapier, Python, OpenAI and Anthropic APIs, RAG, OCR and document AI, REST/webhooks, SQL, CRM and ERP integration, SSO/SAML.Engagement model: a fixed-price teardown of the process first, so you get a written map and an honest verdict before committing to a build. Then deployment. Then I stay and operate what I shipped for 30 days after launch, watching drift, latency and spend against real traffic.I work remote worldwide with deliberate overlap on US, UK and EU hours.