AI Automation Suite — Production Implementation

Six connected n8n workflows — support, sales, finance, BI, HR/IT, and a knowledge-indexing layer underneath all of it — running as one system for a real client, not six separate demos.

Watch the complete system in action — all 6 workflows executing live.

What this project actually is

Look at case studies 01 through 05 above — the support assistant, the lead qualification pipeline, the invoice processor, the executive dashboard, the onboarding system. Each one is documented on its own because each one is a distinct, reusable pattern. This project is what happened when a client needed all five, wired together, plus a sixth piece underneath: a vector knowledge-base indexing layer that keeps the support assistant's answers grounded in the client's actual documentation instead of a static FAQ.

Built in two weeks, self-hosted on n8n rather than a metered cloud platform, and running against the client's real traffic since launch. n8n was the right call here — fast to ship, fully self-hosted, no per-task billing. The same architecture (validation → routing → confidence-based handling → audit trail) carries over just as cleanly to Make, a custom Python/Node service, or a full SaaS backend when that's the better fit for a client's stack or scale — see the Platforms & Stack section for the fuller picture.

Six Production Workflows

01 — Scheduled Analytics

Daily Executive Report

Postgres · Gemini · Gmail

Fetches the last 24 hours of support conversations, computes resolution rate, escalation rate, sentiment, and response time, then emails a structured executive summary at 7am — automatically, every day, including quiet days.

~12s execution time Runs honest — reports real zeros, not silent failures

02 — Document Automation

Intelligent Invoice Processing

Google Vision OCR · Gemini · Drive

Three intake channels feed one pipeline: OCR extracts vendor, amount, date, tax, and line items; the system checks for duplicates before booking, routes for Slack/email approval, and archives the processed file automatically.

~4.3s per invoice Duplicate detection before booking

03 — Webhook Service

Real-Time Customer Support AI

OpenAI/Gemini · PGVector · Slack

A live webhook endpoint that validates the inbound message, classifies intent and confidence, retrieves grounded context from the vector knowledge base, and either resolves automatically or escalates with a written summary attached.

Confidence-routed, not scripted Grounded in the client's actual docs

04 — Sales Automation

AI Lead Qualification & Scoring

HubSpot · WhatsApp · Teams

Scores every inbound lead 0–100 on budget, timeline, industry fit, and urgency, returns a structured recommendation, and syncs straight to HubSpot — with automatic follow-up scheduling for anything that stalls.

100% lead coverage, no manual triage

05 — HR Automation

Employee Onboarding & IT Provisioning

Google Calendar · Sheets · Slack

Builds a role-specific onboarding plan for every new hire — checklist, training path, equipment list — then provisions the HR/IT tasks and notifies every stakeholder automatically, with its own follow-up cadence.

Full personalization by role

06 — Data Pipeline

Vector Knowledge Base Indexing

Gemini Embeddings · PGVector

The layer that makes workflow 03 trustworthy: indexes the client's documentation into a Postgres/PGVector store using Gemini embeddings, so the support assistant's answers are grounded in real source material instead of a static, hand-maintained FAQ.

Powers semantic search for the support AI

Built With Modern AI Stack

Results & Financial Impact

Figures below are client-reported, based on time-tracking comparisons before and after launch.

Production-Grade Architecture, Not a Prototype

Every workflow in the suite shares the same discipline as the case studies above:

• —Automatic retry logic with exponential backoff on every external call
• —Dead-letter queue and audit log for every failed transaction
• —Real-time Slack alerts the moment anything fails
• —Structured, schema-validated AI output — no free-text parsing, no hallucinated fields
• —Webhook token validation and SSL/TLS on every external endpoint

One error workflow, attached to every workflow in the suite — nothing fails silently.

What I Learned Building This

Documentation & Technical Specs

Full write-ups for technical review — these go out on request rather than as public downloads, so I know who's reading them.