Digital Twin · v1.0

I'm Fabrizio's Digital Twin.
Ask me anything

I know Fabrizio's background and curriculum, and I answer in his voice. The real human might just chime in — and I can put you in touch directly.

online · trained on Fabrizio's background
twin@fabrizioamort:~/chat

twin@fabrizioamort initialised. ask about my work, skills, or how to reach the real me.

Enter to send · Shift+Enter for a new lineengineered by me — FastAPI · RAG · SSE on Cloud Run · source ↗

curriculum

Three decades up the stack — now all-in on agentic AI.

~/experience · 30+ years
2024 — now

GenAI Architect

TIM · Telecom Italia · Turin

Architect agentic-AI solutions for enterprise use cases — reference architectures, patterns and guidelines for GenAI. I run multiple PoCs and pilots, some into production.

2007 — 2024

Software Engineer

TIM · Telecom Italia · Turin

Designed and built TIM Group's unified web platform on Drupal — end-to-end responsibility across ~30 enterprise websites: requirements, architecture, development and operations, always for scale and time-to-market.

2003 — 2007

Software Engineer

Shared Service Center · Turin

Built the corporate intranet on SAP Portal in Java, serving roughly 100,000 employees. Where I learned what scale really does to your design choices.

1991 — 2003

Software Developer

Telesoft · Turin

Twelve years of low-level, performance-sensitive work: radiomobile coverage mapping and telephone-switch management. An early, deep respect for systems that must be correct and reliable, not just clever.

Capabilities

Agentic AILLMOpsRAG EvaluationGoogle CloudVertex AI GKE / KubernetesCloud-NativeLangGraph CrewAIMCPPython

Profile

EducationM.Sc. Computer Science — University of Turin
Based inTurin, Italy · remote-friendly
LanguagesItalian (native) · English B2.2

projects

Open source — where I prove the ideas.

github.com/fabrizioamort
multi-agent

Veritasloop

LangGraph · FastAPI · WebSocket · React

An adversarial multi-agent system that verifies news authenticity through structured debate: PRO and CONTRA agents argue up to three rounds before a JUDGE renders one of five nuanced verdicts, streamed live over WebSocket.

agentic RAG

RLM-RAG

Python · OpenAI · sandboxed REPL · Streamlit

A code-writing RAG agent: instead of embedding search, an orchestrator LLM writes Python to explore a prepared document filesystem, executing in a sandboxed two-tier REPL with hard budgets on steps, reads and tokens.

evaluation

RAG Evaluator

FastAPI · DeepEval · ChromaDB · Qdrant · Neo4j

A platform that runs four RAG strategies — vector, hybrid, graph and agentic — on the same test set and compares them with DeepEval metrics, including an explainability view that shows the judge's reasoning on low scores.

this site

Digital Twin

FastAPI · Firestore Vector Search · SSE · Cloud Run

The twin you can chat with above: RAG over my knowledge base, token streaming, abuse guards, and a human-in-the-loop path that pings my phone so the real me can join the conversation live.

about

A senior architect who codes — three decades of building real systems, now pointed entirely at making generative and agentic AI work inside large, regulated enterprises.

I'm a GenAI Architect in Turin with more than 30 years in software engineering and system architecture. My career has climbed every layer of the stack — low-level telecom systems, large web platforms, cloud-native, and now AI — and that long arc is what I value most about how I work today.

I think of myself as a bridge between business, engineering and operations, working end to end: requirements, architecture, framework selection, PoC, evaluation, deployment, and the operational side — observability and governance. My domain is enterprise AI platforms for large organisations, particularly Telco, where long specification cycles, multi-stakeholder alignment, regulation and legacy integration aren't an afterthought — they shape the architecture.

I'm hands-on: I build, evaluate and ship, with opinions formed from real implementations rather than slideware. I'm genuinely excited about LLMs and agents, but I care about evaluation, cost control and knowledge quality as much as the models themselves — because those are what decide whether enterprise AI succeeds or stalls.

— Fabrizio Amort, Turin