AI Automation Specialist · Certified

12 workflows.
One cohesive
automation practice.

From lead capture and RAG chatbots to multi-agent invoice systems, security guardrails, and AI-generated voice briefings — a full-spectrum portfolio built on n8n, spanning data processing, agentic AI, vector search, and web scraping.

12
Workflows built
18+
Tools integrated
5
LLMs deployed
4
Workflow categories
01 · Tech stack
Tools integrated
⚙️
n8n
Orchestration
📋
Fillout
Form capture
🗃️
Airtable
CRM / data store
📊
Google Sheets
Reporting / logging
📄
PDF.co
Document generation
✉️
Gmail
Email / drafts
💬
Slack
Notifications
📅
Monday.com
Task management
🔵
Supabase
Vector store (RAG)
🌲
Pinecone
Vector store (RAG)
🤖
OpenAI
LLM + TTS
🛣️
OpenRouter
Multi-model gateway
🕷️
Apify
Web scraping
🔍
Tavily
Web extraction
🔀
Cohere
Reranker
⚡
Redis
Chat memory
📈
QuickChart
Chart generation
🗂️
Google Drive
File storage
02 · Case studies
12 workflows, 4 categories
Revenue operations
Lead capture · Primary
Lead capture & proposal generation
End-to-end automation from consultation form to branded PDF proposal, CRM entry, and team notification — no manual steps.
Active · Production
Trigger
Customer submits Fillout consultation form selecting a service tier (Basic / Pro / Enterprise).
Purpose
Automate the full sales proposal cycle. Eliminates 30–50 min of manual effort per submission and removes copy-paste errors.
Tools
Filloutn8nAirtableGoogle SheetsPDF.coGmailSlack
Flow
01Form submit
→
02Duplicate check
→
03Route by tier
→
04Write Airtable
→
05Log Sheets
→
06Generate PDF
→
07Send Gmail
→
08Notify Slack
was: 30–50 min
22 seconds
end-to-end processing
was: manual entry
100% automated
CRM + sheet updates
was: inconsistent
Branded PDF
every tier, every time
Duplicate detection
Duplicate detection & smart routing
Pre-write deduplication that prevents dirty CRM data and deterministically routes new vs returning leads with full observability.
Active · Production
Trigger
n8n webhook receives form payload — evaluated before any record is written downstream.
Purpose
Lookup incoming email in Airtable. Route duplicates to update path; new leads to creation path. Both paths notify Slack — nothing dropped silently.
Tools
n8nAirtableSlack
Flow
01Webhook
→
02Lookup email
→
IF foundupdate record
→
IF newcreate record
→
05Notify Slack
Task reporting · Monday.com
Closed task reporting system
Fires on every Monday.com task completion, logs it with manager attribution, auto-counts totals, and pushes a real-time Slack update including a running score.
Active · Event-driven
Trigger
Monday.com webhook fires when a task status moves to Done. Includes challenge-response handshake to activate the integration.
Purpose
Track task completions per manager automatically. Google Sheets auto-counts cumulative totals; Slack delivers the running score in real time. Zero manual reporting.
Tools
Monday.comn8nGoogle SheetsSlack
Flow
01Monday.com webhook
→
02Challenge / event IF
→
03Get item by PulseID
→
04Append done log
→
05Lookup manager count
→
06Slack notification
Slack message includes
Manager · task name · completion time · cumulative total tasks
Data pipeline · Orders
Orders & customer data pipeline
Joins two Google Sheets datasets, deduplicates by Customer_ID, splits by price threshold, summarizes by region/country, and writes to segmented output tabs.
Data pipeline
Trigger
Manual execution. Reads customer.csv and order.csv sheets in parallel from a single Google Sheets workbook.
Purpose
Practice core n8n data-manipulation nodes: Remove Duplicates, IF branch (Price >$50 vs <$50), Merge by Customer_ID, Summarize by Country/Region, field renaming before writes to Total and Less-than-50 sheets.
Tools
n8nGoogle Sheets
Flow
01Get customers + orders
→
02Remove duplicates
→
03IF price >$50
→
04Merge by Customer_ID
→
05Summarize by region
→
06Rename fields
→
07Write sheets
Nodes practiced
removeDuplicatesIFMergeSummarizeSet
AI agents & agentic workflows
Multi-agent · Invoices
Automated invoice reminder & penalty calculator
Two chained AI agents identify overdue invoices, send a manager digest email, update reminder dates, then calculate and write tiered late penalties to Google Sheets — fully autonomously.
Scheduled · Daily
Trigger
Schedule trigger. Both agents share Grok 4.1 Fast (OpenRouter) and a Calculator tool for precision math.
Agent 1 — Last Reminder
Reads all invoices (getData tool), filters Unpaid status, composes a manager digest email, sends via Gmail, then updates the Last Reminder column for every affected row one by one.
Agent 2 — Penalty Calculator
Applies tiered rules: 0.05%/day (1–7 days), 0.1%/day (8–30 days), 0.15%/day (31+ days). Uses Calculator for rounding. Writes Days Overdue, Penalty Rate, Penalty Amount, and Total Amount per invoice.
Tools
OpenRouter (Grok 4.1)Google SheetsGmailCalculator
Flow
01Schedule
→
02Agent 1: getData
→
03Filter unpaid
→
04Send digest email
→
05updateLastReminder ×n
→
06Agent 2: getData
→
07Calculate penalties
→
08updatePenalty ×n
Key concepts
Tool-calling agents, agent chaining, fromAI() overrides, Sheets as data source + output, tiered logic in system prompts.
AI reporting · Voice + chart
Automated AI sales report with voice briefing
Scheduled pipeline that pulls sales data, aggregates with JavaScript, generates a stacked bar chart and an AI voice briefing in parallel, then emails both as attachments.
Scheduled · Timer
Trigger
Schedule trigger reads a Sales Activity Log from Google Sheets — manager, deal status, and amount.
JS aggregation
Groups by manager + status (Done / Lost / Negotiation), sums revenue, outputs structured arrays for QuickChart stacked bar visualization.
AI Agent — Financial Analyst
Receives manager JSON. Applies historical win rate to pipeline amounts ("Real Expected Revenue"). Outputs a 400-char conversational voice script — no markdown, no special chars — formatted for TTS.
Tools
Google SheetsOpenRouter (GPT-4.1)OpenAI TTS-1-HDQuickChart.ioGmail
Flow
01Schedule
→
02Get Sheets data
→
03JS aggregation
→
04aAI Agent → voice script
04bQuickChart → PNG
→
05TTS → .mp3
→
06Merge binary data
→
07Gmail + attachments
Key concepts
Parallel branches, Merge node for binary data, OpenAI TTS-1-HD, Chart.js via HTTP API, Memory Buffer Window, Calculator tool.
RAG & knowledge systems
RAG chatbot · Supabase
Stairlift knowledge base chatbot
Two-phase RAG system: a PDF ingestion pipeline that vectorizes a product brochure into Supabase, and a live chat agent that retrieves and reranks context before answering.
Chat + Slack output
Phase 1 — Ingestion
Downloads Pinnacle-Brochure.pdf from Google Drive → extracts text → Recursive Character Text Splitter → OpenAI embeddings → inserts into Supabase "stairlifts" table.
Phase 2 — Retrieval chat
Chat trigger activates RAG Agent (OpenRouter GPT-4.1 mini + Redis persistent memory). Vector search retrieves top-20 chunks, Cohere reranker scores relevance, LLM generates answer. Response forwarded to Slack DM.
Tools
Google DriveSupabaseOpenAI EmbeddingsCohere RerankerOpenRouterRedisSlack
Retrieval flow
01Chat trigger
→
02RAG Agent
→
03Vector search topK=20
→
04Cohere rerank
→
05LLM response
→
06Slack DM
Key concepts
Two-phase RAG, Cohere reranking for precision, Redis persistent memory, retrieval-as-tool pattern, system prompt persona design.
RAG ingestion · Pinecone
Power.ai document ingestion processor
Form-triggered pipeline that accepts PDF uploads from non-technical team members, extracts text, and pushes embeddings into a named Pinecone namespace — powering the email classifier's RAG responses.
Form trigger · On-demand
Trigger
n8n Form Trigger with labeled "Document Type" field and a PDF file upload — no code required from the uploader.
Purpose
Non-technical staff add internal knowledge to the RAG index without touching n8n. Strips newlines, embeds via OpenAI, stores in Pinecone namespace "power-ai-core-internal-knowledge".
Tools
n8n FormOpenAI EmbeddingsPinecone
Flow
01Form submit
→
02Extract PDF
→
03OpenAI embed
→
04Pinecone insert
Key concepts
Namespace-scoped vector storage, form-as-ingestion-UI, PDF extraction pipeline, stripNewLines for clean embeddings.
Email AI · Classifier + RAG
AI email classifier & auto-draft system
Polls Gmail every minute, classifies each email into 3 categories with LLM, then routes to distinct handling paths — querying Pinecone for business leads, drafting personal replies, or doing nothing for noise.
Active · Every minute
Trigger
Gmail trigger polling every minute. Subject + body passed to LLM Text Classifier (GPT-4.1 mini). Three output branches.
Branch A — Power.ai leads
Labels IMPORTANT → queries Pinecone RAG index for context-aware response → Slack notification (new-inquiries channel) → creates Gmail draft.
Branch B — Private messages
Labels PERSONAL → LLM Chain with personal assistant persona writes a draft → Slack notification → saves Gmail draft.
Branch C — Other
No operation. Noise filtered silently.
Tools
GmailOpenAI GPT-4.1 miniPineconeSlackLLM Chain
Flow
01Gmail poll
→
02LLM classify
→
ALead: label
→
Pinecone query
→
Slack + draft
BPersonal: label
→
LLM draft
→
Slack + draft
COther: no-op
Key concepts
Text Classifier node, LLM Chain persona drafting, Pinecone HTTP API query, Gmail label + draft creation, multi-branch routing.
Security & data acquisition
Security · Defense in depth
Multi-layer security guard for AI agents
Two-layer webhook guard that blocks prompt injections, PII exposure, and jailbreak attempts before any input reaches an AI model — returning HTTP 403 on violations and 200 on clean, sanitized payloads.
Webhook · Real-time
Trigger
POST /security-check webhook. Extracts user_message and timestamps the request on arrival.
Layer 1 — Violation detection
Keyword guardrail (ignore all, jailbreak, system prompt, bypass...), PII detection (credit card, SSN, email, phone), AI jailbreak detection at threshold 0.7. Violations → 403 Blocked immediately.
Layer 2 — Sanitization
Approved messages: PII replaced with [TYPE_REDACTED], URL credentials stripped. JavaScript logs diff of original vs sanitized. Packages clean payload for downstream AI.
Tools
n8n GuardrailsOpenAI GPT-4o miniWebhookJavaScript
Flow
01POST webhook
→
02Extract + timestamp
→
03Layer 1: detect
→
FAILFormat 403
PASSLayer 2: sanitize
→
04Log + package
→
05200 Approved
Key concepts
Defense-in-depth pattern, n8n Guardrails node, keyword + AI jailbreak detection, PII redaction, was_sanitized diff logging, HTTP response codes as security signals.
Lead gen · Apify + AI
Web scraping & AI lead enrichment pipeline
Runs Google Maps Scraper via Apify, filters and deduplicates results, visits each website via Tavily, generates a per-company AI summary, and writes an enriched lead database to Google Sheets.
On-demand · Lead gen
Trigger
Manual trigger. Apify runs Google Maps Scraper for a configured search term + city (e.g. "restaurant" in New York), returning up to 15 results per search.
Enrichment
Each passing lead: email filter → title/phone/address/website completeness filter → Remove Duplicates → Loop per item → Tavily website extract → LLM Chain (3–5 sentence AI summary: specialty, features, hours, address) → Sheets append.
Tools
ApifyTavilyOpenRouter (GPT-5 nano)Google Sheets
Flow
01Apify scrape
→
02Filter emails
→
03Split out
→
04Filter complete
→
05Remove dupes
→
06Loop
→
07Tavily extract
→
08LLM summarize
→
09Sheets append
Nodes taught in this lesson
AggregateSplit OutFilterSwitchLoop Over ItemsEdit Fields
Discovery · Methodology
As-Is process analysis & solution design brief
Structured discovery methodology applied before any build — documenting current state, scoring bottlenecks, and defining measurable success criteria via a Solution Design Brief.
Methodology artifact
Purpose
Prevent fragile automation by understanding the current process before touching n8n. Ensures alignment between client need, business logic, and what gets built — eliminating rework.
Deliverables
Solution Design BriefAs-Is process mapPain point registerROI estimateBuild spec
Discovery steps
01Stakeholder interview
→
02As-Is mapping
→
03Pain point scoring
→
04SDB sign-off
→
05Build brief
03 · AI models
Models deployed across workflows
GPT-4.1 mini
OpenAI via OpenRouter
Email classifier · Stairlift chatbot
GPT-4.1
OpenAI via OpenRouter
Sales report AI agent
GPT-4o mini
OpenAI direct
Security guard jailbreak detection
Grok 4.1 Fast
xAI via OpenRouter
Invoice reminder + penalty agents
GPT-5 nano
OpenAI via OpenRouter
Web scraping lead summarization
TTS-1-HD
OpenAI
Sales report voice briefing
OpenAI Embeddings
OpenAI
Stairlift RAG · Power.ai ingestion
Cohere Reranker
Cohere API
Stairlift retrieval precision
OpenRouter gateway
Multi-model routing
Unified LLM access across workflows
04 · Approach
Build methodology & principles
01
Discover
As-Is analysis, SDB, stakeholder alignment, bottleneck scoring before any build begins.
02
Design
Routing logic, tool selection, data schema, error-path planning, security model.
03
Build
n8n workflow implementation with modular, reusable node and agent architecture.
04
Debug
Error handling, duplicate resolution, edge case coverage, minimal reproducible tests.
05
Optimise
PII protection, node consolidation, performance tuning, ROI documentation.
// ERROR HANDLING
Fail loudly, never silently
Every branch has an error path. Failures route to Slack with context. No record dropped silently — from webhook to vector store.
// PRIVACY
PII by design
Sensitive fields route only to systems that need them. Guardrails sanitize and redact before any LLM sees user input.
// ARCHITECTURE
One trigger, many outputs
Single events fan out to CRM, Sheets, email, Slack, and PDF. No polling unless the use case demands it.
// SECURITY
Defense in depth
Layer 1 detects. Layer 2 sanitizes. No user input reaches an AI model without passing both guardrails.
// ROI FRAMING
Time saved × volume = value
Every workflow is quantified: minutes saved per run × monthly volume × team cost rate. Evidence-based decisions.
// HANDOFF
Built to be understood
Sticky notes, clear node naming, and SDB documentation mean anyone can extend or hand off without tribal knowledge.