Production Systems · Multi-Agent Workflows · n8n Cloud

Production n8n Systems & AI Automation Showcase

A real-world engineering portfolio demonstrating resilient n8n pipelines, semantic vector retrieval (RAG), autonomous email triage with LangChain memory, and multi-channel synchronization across Google Workspace, Telegram, and Slack.

⚙️ n8n Cloud Hybrid 🧠 DeepSeek-V3 Reasoning ⚡ NVIDIA Nemotron 120B Embeddings 📫 Gmail API v2 📱 Telegram Bot API 🗄️ Supabase Vector (pgvector) 🔍 Weaviate Vector DB 📊 Google Sheets API v4 🧩 LangChain Memory
4 Live Production Workflows
< 5s Real-Time AI Response Time
100% Zero-Touch CRM Ingestion
99.9% Resilient Retry & Fault Tolerance
System 01 · Client Acquisition Funnel

Lead Intake & Multi-Turn Strategy AI

🟢 Active Pipeline

A dual-entry qualification pipeline. Leads entering via Telegram Bot or Frontend Webhooks are normalized, evaluated through a multi-turn DeepSeek strategist, conditionally branched, and pushed to notification hubs.

Step 01 · Entry A

Telegram Bot

Instant conversational trigger when a prospective client messages.

Step 01 · Entry B

Webhook POST

Structured payload received from web forms or ad landing pages.

Step 02 · Cleanse

Normalize Input

Extracts unified session_id and user_message across all sources.

Step 03 · Reasoning

Strategy Agent

DeepSeek conversational AI with memory narrow scopes in ≤3 turns.

Step 04 · Logic

JSON Switch

Routes clarifying questions back to user or triggers team alerts.

Lead Intake n8n Workflow Canvas
n8n Workflow Canvas (Click to Enlarge) Multi-channel convergence (Telegram + Webhook) → Normalize Input → DeepSeek Multi-Turn Agent → Conditional Switch.

🧠 The Multi-Turn Strategy Engine

Unlike rigid web forms that drop drop-offs, the DeepSeek AI agent acts as a live qualification consultant:

  • Dynamic Clarification: Asks up to 3 targeted follow-up questions one at a time to determine budget, scope, and urgency.
  • Persistent Memory: Remembers previous dialogue context linked to the user's unique session_id.
  • Strict JSON Handshake: Enforces schema { "is_pass_next": boolean, "message": "..." } for programmatic branching.

⚡ Enterprise Operational Value

Engineered to eliminate manual qualification friction for service and agency operators:

  • Instant Qualification: Sub-5-minute lead qualification when intent is highest.
  • Multi-Channel Adaptability: Easily connects to GHL, Zapier, Google Ads lead hooks, or custom web forms.
  • Automated CRM Record: Pushes sanitized, structured notes directly to sales CRM and Slack channels.
System 02 · Autonomous Operations

Autonomous AI Gmail Triage, Auto-Responder & Multi-Channel CRM

🟢 Active Production · TDABqXYoXrEUSlHi

An intelligent 24/7 email operations hub. Automatically polls incoming messages, routes newsletters and marketing noise to archive, leverages an AI Agent with conversation memory to compose custom replies to high-value inquiries, syncs to a Google Sheets CRM with 5x retry logic, and sends instant Telegram push cards to mobile.

Step 01 · Trigger

Gmail Polling

Fetches unread inbox messages; extracts sender, subject, body, & headers.

Step 02 · Triage

AI Classification

Tags as Work, Personal, Marketing, Google, or Spam; auto-archives noise.

Step 03 · Auto-Reply

Memory AI Agent

Checks history & drafts context-aware replies via Gmail API.

Step 04 · Sync

Sheets CRM (5x Retry)

Appends row with timestamp, sender, subject, label, & message ID.

Step 05 · Alert

Telegram Push

Sends formatted VIP alert directly to your phone within seconds.

⚡ 1 · 24/7 Sub-5-Minute Lead Response

The Problem: Responding within 5 minutes makes you 7x more likely to win the client. If an email arrives overnight or during meetings, leads go cold.
How It Helps: The AI Agent reads the inquiry, accesses conversation history, and sends a professional reply in seconds. Your business appears responsive 24/7.

📥 2 · 2–3 Hours Saved Daily (Inbox Zero)

The Problem: Over 70% of inbox traffic is marketing newsletters, sales pitches, and system alerts that bury high-priority messages.
How It Helps: Marketing emails are instantly archived. Spam is quarantined with dedicated labels. Your primary inbox only displays actionable work.

📊 3 · Zero-Maintenance Google Sheets CRM

The Problem: Manual data entry into CRMs is time-consuming, prone to human typos, and regularly neglected.
How It Helps: Automatically records Sender, Subject, Category, Timestamp, and Message ID. Built with exponential backoff (5x auto-retry) for zero data loss.

🔔 4 · Focused Mobile Push (Telegram)

The Problem: Noisy email notifications break focus and cause screen fatigue during deep work sessions.
How It Helps: Mute noisy Gmail notifications. Your private Telegram bot only pings you when a legitimate Work or Personal email arrives.
💡 Live Operational Scenario: Agency Client Inquiry

Here is the step-by-step automated timeline when a prospect reaches out after business hours:

11:30:00 PM
Prospect emails: "Hi, I need a website redesign and automation quote for my business."
11:30:04 PM
Workflow polls inbox, categorizes as Work, and applies Gmail label.
11:30:08 PM
AI crafts polite reply confirming receipt & setting next-day expectation.
11:30:10 PM
Logged to Sheets CRM & Telegram bot alerts mobile: New Lead: Redesign Inquiry.
System 03 · Retrieval-Augmented Generation (RAG)

Airbnb RAG Guest Concierge — FiFi

🟢 Production Vector Architecture

An autonomous hospitality AI assistant. Features an automated hourly Knowledge Base synchronizer connected to Google Drive and Supabase vector store, coupled with a guest-facing conversational agent equipped with safety guardrails.

1 · Vector Store Auto-Sync Pipeline

Maintains fresh property information without manual re-indexing:

  • Google Drive Poller: Monitors property documentation for updates every 60 minutes.
  • Text Chunker: Splits content into 500-character windows with 50-character semantic overlaps.
  • NVIDIA Nemotron 120B Embeddings: Converts chunks into dense high-dimensional vectors.
  • Supabase (pgvector): Inserts embeddings with cosine distance matching for instant lookup.

2 · RAG Guest AI Concierge

Answers guest questions with property-grounded precision:

  • Property-First Priority: Searches Supabase vector store before falling back to general search.
  • Amenity Direct Links: Provides hot tub, TV, WiFi, and appliance operation guides instantly.
  • Safety Guardrails: Politely declines unauthorized guest additions and routes emergencies to host/911.
  • Audit Trail: Logs guest question, AI answer, and timestamp to Google Sheets.
System 04 · Vertical Market Intelligence

Real Estate Market Trend Report — AI Market Intelligence

🟢 Enterprise Semantic Search

A vertical RAG assistant for real estate professionals. Ingests raw market reports into Weaviate using Cohere embeddings, allowing brokers and investors to query real-time trends, regional prices, and historical performance with zero hallucinations.

1 · Market Data Ingestion Pipeline

Processes market reports, listings, and pricing updates:

  • Custom Ingest Form: Secure endpoint for raw real estate bulletins and listings.
  • Recursive Splitter: 400-character chunking with 40-character overlap for optimal semantic density.
  • Cohere embed-english-v3.0: State-of-the-art embedding model for commercial search.
  • Weaviate Vector DB: Stores vectors in isolated collections for instant filtered querying.

2 · Grounded Q&A + Audit Logging

Ensures verified data delivery to clients and brokers:

  • Grounded Reasoning: If market data lacks relevant stats, the agent prompts for clarification rather than guessing.
  • Conversation Memory: Seamless multi-turn session tracking for complex financial queries.
  • Complete Compliance Log: Records full user queries, responses, and latency in Google Sheets.

Ready to Automate Your Business Operations?

From custom n8n workflow architecture to autonomous AI agents and CRM integrations, let's build reliable systems that save time and close more deals.