A case study by Antoine Moineaux, Business Operations & AI Automation
Project: OpsPilot - AI Lead Intake & Follow-Up System
Value: Inbound demo requests are automatically validated, classified by AI, routed into HubSpot CRM, logged in Airtable, and surfaced to a human operator via Telegram - replacing a manual inbox triage process with a structured, auditable, human-controlled pipeline.
Stack: Tally · n8n · Claude (claude-sonnet-4-6) · HubSpot CRM v3 · Airtable · Telegram
Project type: Built for a real client. The system was designed from a real discovery process, deployed in production, and processed real inbound requests. The client's identity and industry are not disclosed. All screens, data, and business context shown here have been anonymized for public portfolio use.
Antoine’s role: AI Automation Specialist - solo build, end to end: discovery, process mapping, workflow architecture, AI prompt engineering, CRM integration, audit trail design, human-in-the-loop logic, sub-workflow design, end-to-end testing.
Status: Fully functional - validated end-to-end across multiple QA scenarios.
A full end-to-end walkthrough of the OpsPilot pipeline: from Tally submission to AI classification, CRM routing, and the Telegram human-in-the-loop approval.
Small B2B SaaS teams receive inbound requests every day: demo requests, support tickets, partnership inquiries, incomplete submissions, duplicates. Without a structured system, every message requires manual reading, CRM data entry, priority assessment, and follow-up - creating operational overhead that scales poorly and produces inconsistent results.
OpsPilot is a complete AI-powered lead intake system built to solve this problem. A structured Tally form captures inbound requests. n8n orchestrates the full pipeline: validating and normalising submissions, querying HubSpot for duplicates, calling Claude for AI classification, routing based on deterministic business rules, writing to HubSpot and Airtable, and alerting the operator via Telegram with inline Approve / Reject buttons.
The result is a single, traceable intake system where every inbound request is classified and surfaced to a human - no outbound reply is sent without human review, and every completed execution is logged in an independent audit trail.
Note on this case study: OpsPilot was designed from a real operator’s validated requirements and has been running in production since June 2026, currently handling around 100 requests per week. The public version uses anonymized data. The client’s identity and industry are not disclosed.
This case study documents the architecture, the technical decisions, the end-to-end demo walkthrough, and the operational reasoning behind each design choice.