AI Automation & Agents

Software that does the work you keep redoing.

I build AI agents and n8n automations for real business operations — answering emails, qualifying leads, taking calls, and searching your own documents. Every system is designed to escalate to a human when it's unsure, so it never hands a customer a made-up answer.

PROJECT 01

AI Customer Support Agent — auto-drafts replies, escalates the unknown

Associated with Upwork · RAG
Human-in-the-loop

An AI agent that reads incoming customer emails, finds the answer inside the company's own help docs using retrieval-augmented generation (RAG), and drafts a reply automatically.

Stack
n8nGeminiVector embeddingsRAGConfidence routing
The key idea

It scores its own confidence. Sure of the answer → it drafts a reply. Docs don't cover the question → it escalates to a human instead of guessing. It never hallucinates answers to customers, and it cuts response time while keeping a person on the edge cases.

PROJECT 02

AI Lead Capture & Qualification System

Associated with Upwork · n8n · Gemini · HubSpot
24/7

Captures leads from website forms and emails, then qualifies them automatically. AI reads each message and classifies it hot, warm, or cold — with a reason. The lead syncs to HubSpot CRM, logs to Google Sheets with full detail, and the sales team gets an instant Slack alert showing the AI's classification and the lead's intent.

Stack
n8nGoogle GeminiHubSpot CRMGoogle SheetsSlack
The outcome

Replaces slow manual lead sorting. Every lead is captured, scored, and routed in seconds, 24/7 — so no lead goes cold and sales always knows what to chase first.

PROJECT 03

AI Voice Receptionist — answers calls & books appointments

Vapi · n8n · Google Calendar · ElevenLabs
Voice AI

A voice receptionist that answers incoming phone calls, talks naturally like a real person, and handles bookings end to end. It answers common questions, checks availability, books straight into the calendar, captures the caller's details, and transfers to a human when the situation needs one — so no call goes unanswered, 24/7.

Stack
Vapin8nGoogle CalendarElevenLabsGoHighLevel-ready
Live demo

In a recorded call, the agent books an appointment entirely by voice — the booking then appears in the calendar with the lead logged automatically. Demo recording available on request.

PROJECT 04

Multi-Agent System (LangGraph)

LangGraph · StateGraph · MLflow
Advanced

A StateGraph with specialized director, architect, engineer, QA, and scorer nodes, using conditional edge routing and human-in-the-loop middleware, with MLflow logging for every run. The kind of orchestration behind serious, auditable agent workflows.

Stack
LangGraphConditional routingHITL middlewareMLflow
Why it matters

Most "AI agents" are a single prompt. This is a coordinated team of agents that can plan, build, check, and score their own work — with a human able to step in at any node.

PROJECT 05

RAG Knowledge-Base Chatbot

Pinecone · Google Drive · Conversation memory
RAG

Auto-indexes Google Drive documents into Pinecone and answers employee questions straight from company files — with conversation memory so follow-up questions make sense. Turns a messy shared drive into something you can just ask.

Stack
PineconeGoogle DriveVector searchMemory
Good for

Teams drowning in docs, SOPs, and policies — onboarding, internal support, and "where's that file?" questions answered instantly.

The rule I build by

An agent that guesses is worse than no agent at all.

Every system I ship has a confidence gate and a human escalation path. When the AI knows, it acts. When it doesn't, it hands off — quietly, before a customer ever sees a wrong answer. That's the difference between a demo and something you can actually put in front of clients.