AI Automation Engineer
I build automations that qualify leads, respond to customers, extract data from documents and update your CRM — with error handling, retries and human fallback built in.
Beyond the prompt
A ChatGPT subscription won't update your CRM, handle duplicates, retry after an API error or route requests your AI can't handle to your team. I build that part.
What I can do for you
Conversational assistants on WhatsApp or web widget that answer using your knowledge base, qualify leads, collect data with consent and hand off to a human when needed.
End-to-end automations between forms, HubSpot, Google Sheets, email and external APIs. AI where it actually helps, not everywhere by default.
Structured data extraction from invoices, orders and PDFs into your CRM, spreadsheet or accounting system. Schema validation, confidence scoring and manual review for edge cases.
Beyond the happy path
The value isn't making a workflow run once. It's making it run on Tuesday morning when the API returns 429, the lead is a duplicate and the PDF is scanned sideways.
Payload validation, HTTPS, credentials in environment variables, minimum-scope tokens, input sanitization.
Backoff retries for timeouts and 429/5xx, dead-letter queue for failed events, email or Slack alerts.
Execution ID, process state (received, validated, processed, failed), minimal payload, zero sensitive data in logs.
Manual review for low confidence, ambiguous AI output, critical requests or cases the AI shouldn't handle alone.
About me
I'm Luigi Scorzelli. I've spent the past 5+ years building web applications — frontend, backend, APIs, databases, deployments. Over the last year I took all that experience and applied it to AI engineering: LLM agents, n8n workflows, RAG architectures and CRM integrations.
The result: I can build the entire chain, from the form the user fills out to the record that appears in your CRM, including the AI that qualifies, validates and routes. Not just the "smart" part, but everything that makes it reliable.
Vue, Angular, Laravel, Node.js, Python, REST APIs, MySQL, Docker, CI/CD.
RAG, pgvector, embeddings, function calling, structured outputs, guardrails.
CRM pipelines, contact/deal/task, webhooks, dedup, automated emails.
I work with Italian and international clients, written communication and video calls.
Proof of work
Each project simulates a realistic client and shows the full flow: from trigger to CRM output, with error handling, security and fallback documented.
A single system for a dental clinic, with two connected flows. The patient writes on WhatsApp or the website widget and gets answers grounded only in the clinic's knowledge base — hours, services, prices, insurance — with conversational memory and human handoff on emergencies and out-of-scope cases. When they fill out the appointment form, in under 30 seconds the AI qualifies the lead (urgency, service, value) and writes to HubSpot: Contact, Deal, Task and Note, with an automatic confirmation email.
NordFit Store, a fictional fitness e-commerce. The agent recommends setups based on goal, space, level and budget. Users can also interact by voice: the widget supports speech input, catalog and policy answers, purchase-intent qualification and HubSpot CRM preparation.
Upload invoices, orders or documents. AI extracts structured data with schema validation and confidence scoring. Clear data goes to the system. Ambiguous data goes to manual review.
Luna Blu Retreat, a hospitality for direct requests. The booking engine checks dates, availability, capacity, minimum stay, extras and estimated price; n8n generates an AI summary, updates HubSpot and sends the confirmation email.
Technical skills
I combine full-stack experience with AI tools to build the entire chain: from the user interface to the record in your CRM.
AI Agents, LLMs, RAG, function calling, structured outputs, multi-agent systems, guardrails, prompt injection mitigation.
n8n, LangChain, LangGraph, OpenAI API, Claude, embeddings, pgvector, Chroma.
Vue.js, Angular, JavaScript, TypeScript, Laravel, Node.js, Python, FastAPI, REST APIs, MySQL, PostgreSQL.
Docker, GitHub Actions, Google Cloud Run, HubSpot API, Google Sheets API, WhatsApp Business API, SMTP.
How I work
Clear scope, quick demo, measurable delivery. No scope creep, no surprises.
Show me how you work today: where you lose time, where things break, what tools you already use.
I build a working prototype with test data. If it doesn't convince you, we stop here.
I integrate your CRM, APIs, database, email and real channels. I add error handling, logs and security.
Deployment, documentation, video walkthrough. You know exactly what the system does and how to modify it.
Frequently asked questions
A chatbot just answers. An automation connected to your CRM turns the conversation into finished work: it qualifies the contact, opens the deal, creates the follow-up task and sends the email — without anyone on your team re-typing data. The value isn't the answer, it's the process that runs on its own.
Not in my systems: the assistant only answers based on your documents and your catalog (RAG). If the answer isn't in the knowledge base, it says so and hands the request to a human. Every project includes specific tests against hallucinated answers and prompt injection attempts.
That's designed for, not an emergency. If an external service fails, the system retries on its own; if the problem persists, the request goes to a recovery queue and you get an alert. No lead or request is ever lost, and every event stays traceable in the logs.
It depends on scope: a single workflow (e.g. lead qualification from form to CRM) costs far less than a multi-channel system with a knowledge base and a conversational assistant. I work with a fixed quote defined after a free call: you know upfront what you spend and what you get, no surprises.
My process includes a demo with mock data within the first few days: you see it running before committing to the full project. A single workflow ships to production in 1–2 weeks; larger systems take a few weeks more, with visible intermediate deliveries.
Credentials live in protected environments and never in code, connections are encrypted, logs don't store unnecessary sensitive data, and system access uses the minimum required permissions. For highly confidential data, an on-premise architecture with no cloud calls is an option.
Let's work together
Tell me what you want to automate. I'll tell you if I can help, how long it takes and how it would work. No commitment, no blind quotes.
Prefer email? Reach me at luigi.scorzelli87@gmail.com