Field Deployment Kit v1.0 2026-06-20

AI Workflow Starter Kit

Two production-ready systems for a Global FAE: automated content quality pipeline and cross-region knowledge base with predictive capability.

Mission Brief

You want two things working together. First, a content pipeline that catches mistakes before your audience does: your draft goes through Claude for refinement, Gemini for independent critique, automated scoring for quality gating, and Slack for your final approval. Second, a knowledge base that turns years of multi-region FAE experience into a prediction engine — so when a new German military client shows up with the same hardware platform your Japanese client deployed last year, the system tells you what they'll need next before they ask.

Below: the n8n workflow file ready to import, the knowledge base architecture with a concrete data schema, and a three-phase deployment roadmap that starts producing value in one week.

SYS-A

Content Quality Pipeline

Four nodes, one principle: no content reaches your audience without a second model's critique and your explicit approval.

Step 1

Claude

Refine draft

→

Step 2

Gemini

Critique & score

→

Step 3

Evaluation

Pass / fail gate

→

Step 4

You

Approve in Slack

Install n8n

n8n runs on your laptop or in the cloud. Pick one:

# Option A: Desktop app (easiest)
Download from https://n8n.io/get-started

# Option B: Docker (if you have Docker)
docker run -it --rm -p 5678:5678 n8nio/n8n

# Option C: Cloud (no install)
Sign up at https://app.n8n.cloud

Import the workflow

Download the workflow file, then open n8n and go to Workflows → Import from File. All nodes and connections are pre-wired.

Download llm-judge-workflow.json

Configure API keys

Open the "Configure Content Input" node and replace the placeholder values:

Field	Where to get it
anthropic_api_key	console.anthropic.com → API Keys → Create Key
gemini_api_key	aistudio.google.com → Get API Key
slack_webhook_url	api.slack.com → Your Apps → Incoming Webhooks → Add New

Corporate environment tip

If your company blocks API access, use n8n's credential store or environment variables instead of pasting keys into the Set node. Keys in the Set node are visible in execution history — for production use, prefer n8n's built-in secrets management.

Test with real content

Paste a LinkedIn draft or presentation outline into the content_input field. Set content_type to match (LinkedIn post, presentation, technical document). Hit Execute Workflow.

Gemini scores across six dimensions: technical accuracy, clarity, audience fit, originality, actionability, and brand safety. The default pass threshold is 80/100 — adjust quality_threshold in the Set node to match your standards.

Review in Slack

The workflow sends a formatted Slack message with scores, critique, and the final draft. Content that passes the threshold arrives as "Ready for Approval" with the full text. Content below threshold arrives as "Needs Revision" with specific issues and suggested revisions. You read the message and decide whether to publish — no Slack app setup needed, just an incoming webhook.

SYS-B

Cross-Region Knowledge Base

A two-layer system: structured case management on top, semantic intelligence underneath. The first layer captures what happened; the second predicts what will happen next.

Architecture Layer 1: Case Management Layer 2: Intelligence ┌─────────────────────────┐ ┌──────────────────────────┐ │ Notion (cloud, collab) │ │ AnythingLLM (local, safe)│ │ or │───export──▶│ or │ │ Obsidian (local, secure)│ │ n8n + Pinecone + Claude │ └─────────────────────────┘ └──────────────────────────┘ Ground truth data Semantic search + predictions

Security-first decision

Your clients include military and defense. Sensitive case data should never leave your machine. Use Obsidian + AnythingLLM Desktop for classified content (100% local). Use Notion + cloud LLM only for de-identified general cases.

Tool	Type	Best for	Trade-off
Notion	Cloud	De-identified enterprise cases, team collaboration, relational databases with views	SaaS — some defense orgs block the domain entirely
Obsidian	Local Rec	Military/classified cases, offline use, full data sovereignty	No native collaboration; sync via Git or secure cloud drive
AnythingLLM	Local Rec	Semantic search over your knowledge base without data leaving your laptop	Relies on local compute; complex automations need n8n on top
NotebookLM	Cloud	Quick cross-document analysis, zero setup	Google cloud storage; not for classified data
n8n + Pinecone	Cloud	Automated weekly prediction reports, writing results back to Notion	Requires API keys and ongoing costs; best for Phase 3

Data Schema

Every customer case follows this structure. The consistency is what makes cross-region prediction possible.

Customer Profile

Customer_ID string Codename, not real name. Example: DE-MIL-03

Region enum TW | JP | NL | DE

Vertical enum Military | Industrial | Enterprise

Security_Level enum Unclassified | Confidential | Restricted

Hardware_Platform string e.g. Gen4-Tactical-Server

Compliance_Required string[] MIL-STD-810H, IEC-62443, CE

Enabled_Features string[] Currently active features: [A, B, C]

Case Study Record

Case_ID string e.g. CASE-2026-004

Customer relation Links to Customer Profile

Symptom text What happened. e.g. "G-sensor calibration drift at -20°C startup"

Root_Cause text Why it happened. e.g. "Crystal oscillator below operating temp, PLL unlocked"

Resolution text What fixed it. e.g. "Bootloader update: 150ms delay for oscillator stabilization"

Deployment_Profile_Tag string[] Auto-generated feature tags: [High-Vibration, Low-Temp, Secure-Boot]

Prediction Fields

Predicted_Gaps text Issues this client hasn't hit yet but likely will, based on similar profiles

Next_Best_Action text What to proactively recommend. e.g. "Suggest Feature D + E (low-temp compensation firmware)"

How prediction works in practice

Scenario

A new German military client (DE-MIL-03) deploys Gen4-Tactical-Server with MIL-STD-810H compliance, features A and B enabled.

Your knowledge base contains records showing that JP-MIL-02 and NL-MIL-05 — same hardware, same compliance standard — both activated Feature D (auto temperature compensation) within two months of deployment due to environmental temperature variance.

You ask the system: "Based on similar deployment profiles, what will DE-MIL-03 need in the next 3 months?"

The system identifies the match and recommends: proactively propose Feature D and Feature E before the client encounters the problem. You arrive at the meeting with a solution for a problem they didn't know they had.

SYS-C

Deployment Roadmap

Phase 1 — Week 1

Structured capture

Choose Layer 1 tool (Notion or Obsidian)
Set up schema template from above
Manually enter 5–10 historical cross-region cases
Import n8n workflow, configure API keys
Run first content through the pipeline

Phase 2 — Month 1

Semantic intelligence

Install AnythingLLM Desktop
Import de-identified cases as workspace data source
Write FAE-specific system prompt for querying
Test cross-region case retrieval queries
Iterate on deployment profile tags

Phase 3 — Month 3

Automated prediction

Add n8n workflow for weekly knowledge base scans
Auto-detect new customer deployments
Generate prediction reports via LLM
Send reports to email or write back to Notion
Refine prediction accuracy with feedback loop

REF

Quick Reference

Item	Cost	Note
n8n Desktop	Free	Self-hosted, unlimited workflows
n8n Cloud	€20/mo starter	No install, managed hosting
Claude API	~$0.01–0.03 per run	Sonnet 4.6; $3/$15 per MTok input/output
Gemini API	Free tier available	Gemini 2.5 Flash, generous free quota
AnythingLLM	Free	Desktop app, runs on your laptop
Obsidian	Free	Personal use; paid for commercial
Notion	Free / $10/mo	Free plan sufficient for solo use

Model provider independence

The n8n workflow uses raw HTTP requests, not built-in AI nodes. This means you can swap Claude for OpenAI, Gemini for another reviewer, or even a local Ollama model — just change the URL and payload shape in the HTTP Request node. No workflow rebuild needed.