Hi Louis,
Regarding the 33-agent BI build, my main focus would be on the financial data layer first. Reliable QuickBooks and Sage Intacct syncs with solid idempotency, reconciliation, and failure handling are essential before any higher-level synthesis can be trusted.
I’ve built similar systems before, such as one that integrated Amazon seller data into bookkeeping platforms for VAT and financial reporting automation, as well as a multi-tenant SaaS with real-time analytics.
I’m open to starting Phase 1 with the initial four agents on a fixed price so you can assess the work. Could you share more about your expectations for the L3 Opus tier?
Hi Louis — your 31-agent Claude API + n8n build stood out because it sounds like a real production system, not a demo.
The parts that fit my background best are the ones most builders skip:
- self-hosted n8n workflows that need to stay understandable after handoff,
- Claude API flows with different reasoning/cost tiers,
- source-verification steps before analysis,
- credential-safe integrations across business systems,
- documented operator workflows instead of “it works on my machine.”
Portfolio / live work:
• BuddyBuilds — production automation run logs: https://buddybuilds.netlify.app
• Pickleball Connect — live pickleball resource hub with n8n-approved weekly briefing: https://pickleball-connect.netlify.app
• Hagere Today — Ethiopia/diaspora media site with Birr tools, newsletter, and story submission: https://hageretoday.netlify.app
The BuddyBuilds proof page is most relevant here. It shows approval-gated automations and run-log style examples with the non-happy paths included: approve, decline, timeout, and escalation. That same thinking applies to a multi-agent intelligence system — each agent needs clear inputs, outputs, failure handling, and handoff notes.
If helpful, I can start by taking one of the Phase 1 agents and sending back a build outline: n8n workflow shape, Claude prompt/API boundaries, credentials needed, error paths, and what I’d document for an operator.
I’m based in Salt Lake City, Utah, USA, async-first, and comfortable with careful production handoffs.
— Buddy
Salt Lake City, Utah, USA
https://www.linkedin.com/in/habesha-sew-3b5438321/
Hi,
“Production Claude API experience, not demos” is the line I am answering. I run a fleet of autonomous agents that does client work every day: 627 delivered tasks across 11 projects, 461 of them in the last month, each one passing a machine-checked acceptance step before it counts as done. That acceptance loop is exactly what a 31-agent system needs, otherwise L3 synthesis quietly drifts and nobody notices for weeks.
Relevant to your stack: Python on cron next to n8n self-hosted, Claude API with model split by layer, OAuth with refresh tokens for accounting systems, signed webhooks, retry and recovery after provider outages, credentials kept out of workflow JSON.
I work async and document handoffs, remote from Vietnam (GMT+7). Rate 35-40 USD/hour, or from 1,500 USD for a scoped build.
Happy to show the acceptance harness that decides when an agent is actually done.