Skip to content
Index
WORKFLOW AUTOMATION · LOCAL-RUNNABLE PROTOTYPES · PORTFOLIO PROOF

Automation Kits

Five runnable n8n workflows for operations teams: ticket triage, lead scoring, review response, customer onboarding, and an internal standard operating procedure (SOP) bot. Each runs end to end on a local model and routes only the high-stakes cases to a person. Portfolio proof for the Human-in-the-Loop Automation service — not a separate product SKU.

Role
Sole builder: design, prompts, n8n workflows, human-in-the-loop (HIL) gates
Domain
Operations automation (hospitality-leaning)
Surface
5 kits on one shared, runnable 5-node spine
Stack
n8n · local Ollama (phi4:14b) · JSON mode · rule-based gate
Principle
AI drafts. A deterministic gate escalates the risky cases
Status
Runnable + self-tested end to end on sample runs · production channels pending
The shared workflow architecture: trigger, load inputs, local LLM, and a rule-based human-in-the-loop gate that splits routine cases (auto-resolve) from high-stakes cases (human approval) before a structured result. Below, each of the five kits is shown with a verified sample input, the model's verdict, and which way the gate routed it.
The shared spine, the auto-resolve vs. human-approval split, and each kit's verified sample run.
01

The Operational Problem

Operations teams run on inbound volume: support tickets, sales leads, public reviews, new-account onboarding, staff policy questions. Most of it is routine and could be handled in seconds. A small slice is genuinely high-stakes: a refund dispute, a safety incident, an enterprise contract, a one-star review naming a guest.

The naive fix, letting AI answer everything, is unsafe, because a confident wrong answer lands exactly on that high-stakes slice. The opposite, making a person read everything, spends the scarce resource on cases that never needed it. These kits are built around that tension: absorb the routine volume automatically, and guarantee the high-stakes cases reach a human.

02

The Product Principle

One rule holds across all five: AI drafts, a human approves anything risky, never the reverse. And the escalation decision is not left to the model's discretion alone. Each workflow runs a two-layer gate. First the model returns its own requires_human_approval flag inside strict JSON. Then deterministic code (a fixed rule in plain code, the same input always gives the same result) re-checks the case against kit-specific rules and can force escalation regardless of what the model said.

The model is allowed to be cautious. It is never the only thing standing between an automated action and a guest, a dollar, or a safety event.

03

How One Workflow Runs

Every kit is the same runnable spine of five n8n nodes, no cloud API keys, executable on a laptop:

  • Trigger: a sample case enters the workflow.
  • Load Inputs: the case is paired with the kit's system prompt.
  • Local AI model (a large language model, or LLM): Ollama (phi4:14b) returns strict JSON (json mode, temperature 0.2).
  • Parse + HIL Gate: the JSON is parsed, then the kit-specific rule decides auto-resolve vs. human approval.
  • Result: a structured object plus the human-approval flag.

The same five kits also ship a design-exact production graph that swaps the local model for real channels and models, covered below.

04

The Five Kits

Five kits, one spine. Each classifies, scores, or drafts in its domain and carries its own escalation rule. Every example below is a verified sample run from the local build.

Hospitality Ticket Triage

Human
Decides
category, priority (low → urgent), department, guest-facing draft reply
Escalates when
priority is high/urgent, or the text mentions refund, chargeback, passport, safety, injury, or legal
Verified run
“$2,400 chargeback”priority: urgentHuman

Lead Qualification + Scoring

Human
Decides
how well a lead fits your ideal customer (ICP), 0 to 100, tier A/B/C, intent, next-best action, reply
Escalates when
score ≥ 80, tier A, enterprise/security/legal terms, or a borderline lead with hot intent
Verified run
“VP, multi-property, budget approved”fit score 85Human

Review Response Agent

Auto
Decides
sentiment, category, severity, brand-voice public response
Escalates when
negative sentiment or high severity
Verified run
“5-star review”positiveAuto

Customer Onboarding Engine

Auto
Decides
segment, personalization plan, first steps, risk flags, welcome
Escalates when
enterprise/high-value, single sign-on (SSO), security/legal, money-touching, or missing consent
Verified run
“SMB, non-technical”routineAuto

Internal Ops SOP Bot

Human
Decides
answer, cited SOP references, confidence, reply grounded in approved source documents (retrieval-augmented generation, or RAG)
Escalates when
low confidence, or the question touches a spill, safety, HR, refund, or waiver
Verified run
“Refund policy?”cites P1Human
05

From Prototype to Production

Alongside each runnable kit is a design-exact production scaffold: an importable n8n graph that is not yet run with live data. They share a common shape:

  • Multi-channel intake: webhook, email, schedule, or Telegram.
  • Normalize: dedup, canonical schema, and consent / lawful-basis checks.
  • Cheap classifier → premium drafter: cost-tiered models, with citation-enforced retrieval (RAG) where there is a knowledge base.
  • Force-HIL validation: the deterministic gate, in code.
  • Mandatory approval: a send-and-wait step in Slack or email, with an approve / redraft / escalate router. Execution only fires after approval.
  • Audit + recovery: an immutable log with cost tracking, a dedicated error sub-workflow, and feedback capture for continuous improvement.

Productionizing a kit means swapping the local-Ollama node for its real channels and model, adding credentials, and feeding sanitized data.

Captured walkthrough · 27sReal screen capture
Screen capture of the automation showcase site: the shared intake, classify, human-in-the-loop gate, act pattern and the five workflow kits, each shown with its verified sample runs. This is a demo surface for the kits, run end to end on a local LLM with synthetic sample data. Not a live production system with real customer data.
Scope & Honesty
  • The five runnable kits execute end to end on local Ollama with no cloud API keys, verified on built-in sample inputs (2026-06-02, gate retest 2026-06-19).
  • The production graphs are design-exact and importable, but have not yet been run with real channels or data.
  • No client metrics or ROI claims: this is sample-set proof of the classify/score/draft plus human-gate logic.
  • Sample inputs are synthetic and the embedded SOP policies are illustrative.