Skip to content
View fvalerii's full-sized avatar
  • Self employed
  • Panama

Block or report fvalerii

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
fvalerii/README.md

Hi there, I'm Fabrizio Valerii 👋

🤖 AI Engineer & Agentic Systems Architect

I am a highly analytical AI Engineer specializing in Generative AI, Agentic Workflows (MCP, Multi-Agent Orchestration), and Deep Learning. My professional edge is built on the intersection of rigorous technical research and a multiyear corporate career in Strategic Consulting and Global Finance (Citibank, UniCredit, PwC).

Over the last 2.5 years of intensive engineering immersion, I have transitioned from evaluating strategic equity to architecting production-ready AI systems. I don't just build models; I design deterministic, fault-tolerant pipelines where technical precision meets business governance.

📫 How to Reach Me


🛠️ My Research & Engineering Stack

  • Agentic & Generative AI: LangGraph, CrewAI, BeeAI, AG2, Model Context Protocol (MCP), Large Language Models (LLMs), Diffusion Models, IBM Watsonx.
  • Production RAG & Data Prep: LangChain, LlamaIndex, Docling, MultiModal Processing, Knowledge Graphs (Neo4j), Vector Databases (ChromaDB, FAISS).
  • Computer Vision: Vision Transformers (ViT), Hybrid CNN-ViT Architectures, Variational Autoencoders (VAEs), Image Processing (OpenCV).
  • Deep Learning & Probabilistic: PyTorch, TensorFlow/Keras, TensorFlow Probability, Normalizing Flows, Bayesian Neural Networks, Uncertainty Quantification.
  • MLOps & Cloud: Python (Expert), SQL (BigQuery), Vertex AI Pipelines, Kubeflow, Google Cloud Platform (GCP).

##🚀 High-Impact Portfolio & Hackathons

1. TireForge Predictive Maintenance — Multi-Agent AI System (Jun–Jul 2026)

  • Associated With: Agent Master - Digital Agentic AI Hackathon (Founderz)
  • Core Logic: Real-time multi-agent orchestration built on Azure AI Foundry.
  • Achievements: Engineered a production-grade system to automate industrial fault diagnosis. Features Agentic RAG grounded in official equipment manuals, dynamic external database queries for live spare-part inventory, and a strict Human-in-the-Loop (HITL) approval gate before autonomously opening CMMS work orders.

2. Multi-Modal Evidence Review — Agentic Vision Pipeline (Jun 2026)

  • Associated With: HackerRank Orchestrate (24-hour Hackathon)
  • Core Logic: Deterministic, state-graphed agentic vision pipeline.
  • Achievements: Built a production-minded system to verify insurance-style damage claims by synthesizing images, chat logs, and user history. Designed with strict Pydantic schemas, safety guardrails, provider failover mechanisms, and a mocked end-to-end evaluation suite running offline in $\sim 2$ seconds.

3. Agentic Q&A System: LangGraph & Multi-Agent Orchestration

  • Core Logic: Self-correcting retrieval workflow with Multi-Tenant Isolation.
  • Achievements: Architected a pipeline utilizing IBM Watsonx and a Granite-4 verification loop to eliminate hallucinations. Implemented Docling and hybrid retrieval (BM25 + ChromaDB) to automate high-fidelity technical document synthesis.

4. Geospatial Land Classification: CNN & ViT Hybrid Study

  • Core Logic: Benchmarking Framework Parity (PyTorch vs. TensorFlow).
  • Achievements: Integrated CNN feature extractors with Transformer self-attention blocks to capture global spatial relations. Achieved >99% accuracy and 1.000 ROC-AUC across both framework implementations.

5. Probabilistic Models: VAE & Normalizing Flows

  • Core Logic: Generative Latent Space Organization.
  • Achievements: Used Normalizing Flows for custom data generation and a VAE with $\beta$-weighting to force clear latent-space organization. Validated performance with a 0.4473 FID score, proving near-identical statistical distribution to real data.

##🎓 Certifications & Advanced Academic Research

Engineering & Mathematics

  • MITx MicroMasters in Statistics and Data Science (In Progress, MIT/edX)
    • Completed: Probability - The Science of Uncertainty (6.431x), Machine Learning with Python (6.86x), and Learning Time Series with Interventions (IDS.S24x).
  • STATSX0001: Statistical Learning (Stanford Online)
  • MSc in Management Engineering (Politecnico di Milano)
  • MITx Micromasters in Statistics and Data Science (In Progress, MIT/edX)
    • Completed: Probability - The Science of Uncertainty (6.431x), Machine Learning with Python (6.86x), Learning Time Series with Interventions (IDS.S24x).

Professional AI Qualifications (2024–2026)

  • Model Context Protocol (MCP) Mastery (Anthropic/Fractal Analysis)
  • Building Diffusion Models (Fractal Analysis)
  • IBM RAG & Agentic AI Professional Certificate
  • Google Cloud Professional Machine Learning Engineer (Preparation Completed)
  • IBM GenAI Engineering Professional Certificate
  • IBM Deep Learning Professional Certificate
  • TensorFlow 2 for Deep Learning Specialization (Imperial College London)
  • Google Advanced Data Analytics Professional Certificate

💼 Prior Corporate Leadership

Before dedicating my career to AI architecture, I spent over two decades driving operational excellence and capital allocation in the financial sector:

  • Chairman and Owner, Strategic Project Overseas Inc.: Oversaw private equity investments in young technology companies, focusing on strategic valuation and operational due diligence.
  • Equity Trader (Independent): Developed and executed proprietary investment strategies based on technical and fundamental analysis of financial markets.
  • Organization Manager, Pioneer Global Asset Management: Coordinated large-scale business rationalization, process optimization, and project management (PRINCE2) for Asset Management division of the UniCredit Group.
  • Senior Business Consultant, PWC Consulting: Participated in projects in the financial sector in the areas of strategy, company restructuring, and implementation of IT systems.
  • Telephone Banking Head, Citibank: Implemented the Telephone Banking unit of Citibank in Italy from the ground up and subsequently managing its operations.

🌍 Beyond the Terminal

Polyglot: Italian (Native), English (Advanced), Spanish (Advanced), Portuguese, French, German.

When I am not optimizing transformer workflows or writing Python scripts in Cursor, I manage Cats of Panama, an international animal rescue and adoption platform.

Pinned Loading

  1. waze-churn-prediction waze-churn-prediction Public

    Waze User Churn Analysis: Binary classification using Random Forest & XGBoost. Implements advanced feature importance auditing, hyperparameter tuning, and decision threshold optimization for high-c…

    Jupyter Notebook

  2. nmt-seq2seq-translation nmt-seq2seq-translation Public

    🚀 High-performance NMT study scaling Seq2Seq LSTMs to 200k+ sentence pairs. Features a streaming tf.data pipeline, Transfer Learning (NNLM), and masked loss. Reaches 17.32 BLEU on English-to-German…

    Jupyter Notebook