AI Engineer · Matteo Calza Metre
Agentic AI systems, built for production.
I'm Matteo, an AI Engineer at Formula. I design and ship agentic AI systems and LLM applications for enterprise software, from retrieval and orchestration to evaluation and LLMOps.
LLM systems are software. They have to be measured, constrained and traceable.
AI Engineer at Formula, specialized in agentic AI systems and LLM applications for enterprise software.
I work on an agentic Text-to-SQL assistant that plugs into customers' ERP systems and lets users query company data in natural language. My core expertise is agent orchestration, GraphRAG retrieval, deterministic guardrails, LLM evaluation and LLMOps, deployed on-premise with open-source models.
What I bring to the table
The areas I work in every day, and the tools I use to ship them.
Agentic systems
Agents that plan, call tools and correct themselves under deterministic guardrails.
- LangGraph
- LangChain
- Tool calling
RAG & GraphRAG
Retrieval over documents and databases, from chunking and hybrid search to reranking and graphs.
- ChromaDB
- Memgraph
- MongoDB
- Cohere Rerank
LLM evaluation & LLMOps
Systems that are measured and traceable: custom metrics, gold datasets, tracing, reproducible runs.
- Langfuse
- vLLM
- llama.cpp
- MinIO
Production engineering
From prototype to deployed services, on-premise with open models or on Azure.
- Python
- FastAPI
- Docker
- Azure OpenAI
Experience
AI Engineer
Jul 2026 – PresentFormula SpANow
Agentic AI and LLM applications for enterprise software, on-premise with open-source models.
AI Engineer
Jun 2025 – Jul 2026EY
RAG agents and enterprise chatbots over contracts and corporate documents, with GPT-4o and Azure.
Data Analyst
Sep 2023 – Jun 2025EssilorLuxottica
Power BI dashboards and SQL analysis of production data for quality and scrap reduction.
Latest articles
Notes from building LLM systems for the enterprise: agents, retrieval, evaluation and LLMOps.
AI Agent Observability in Production: How to Trace an Enterprise Agent with Self-Hosted Langfuse
What AI agent observability is, why agentic components need it, what tracing a modern agent actually means, and how I traced an on-premise enterprise agent with self-hosted Langfuse and LangGraph.
- Agents
- Observability
- LLMOps
How to Evaluate an Agentic Text-to-SQL System in Production
Execution accuracy gives one number to very different failures. How I evaluated an agentic text-to-SQL system in production with independent metric families: retrieval, output and SQL components, never folded into a single score.
- Agents
- Evaluation
- Text-to-SQL
How to Make LLM-Generated SQL Safe to Run in Production
A system prompt is a request, not a contract. Here's the deterministic guard that parses every query an agent writes and refuses anything that isn't a safe read, before it reaches a production ERP.
- Agents
- Guardrails
- Text-to-SQL
Get in touch
Open to conversations about agentic AI, LLM evaluation and LLMOps in the enterprise. Reach out on any channel below, or send a message with the form. I usually reply within a couple of days.
Send a message
Fill out the form and I'll get back to you soon.