Skip to content
View albertobarnabo's full-sized avatar
11 followers · 11 following
  • European Central Bank
  • Frankfurt, Germany

Achievements

Achievement: Pull SharkAchievement: YOLOAchievement: StarstruckAchievement: Pair Extraordinaire

Achievements

Achievement: Pull SharkAchievement: YOLOAchievement: StarstruckAchievement: Pair Extraordinaire

Block or report albertobarnabo

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

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

Report abuse
albertobarnabo/README.md

Alberto Barnabò

Double M.Sc. in Computer Science Engineering

Portfolio Website • LinkedIn • Hugging Face

Applied AI Engineer @ European Central Bank

  • Production LLM Pipelines: Designed and deployed end-to-end LLM pipelines on AWS for large-scale regulatory document processing, optimizing context construction and tool-use strategies to improve agent reliability in production.
  • Agentic Workflows: Iterating on execution loop designs and model-facing improvements for long-horizon document analysis—bridging the gap between brittle prototypes and production-grade behavior.
  • Business-to-AI Translation: Transforming open-ended institutional challenges into tangible, usable AI applications that deliver immediate value to non-technical stakeholders.

Open-Source Work

Project What it is Highlights
E-commerce product search · HF Self-hosted semantic search: a retriever and a cross-encoder reranker fine-tuned on Amazon ESCI, running on CPU with no per-query fees 427k human judgments · nDCG@10 0.748 · 2,100 q/s
Fiduciary · HF Local-first financial advisor: Qwen3-4B LoRA on Apple MLX with a tool-using agent loop — nothing leaves the machine 8.7k+ downloads · fused MLX, adapter and GGUF
BurocrazIA · Dataset The first benchmark for Italian document AI: a 104-field schema, gold-annotated documents, a hallucination-counting scorer 204 gold docs · Qwen2.5-VL leaderboard
Synthetic receipts for OCR · Dataset 32,000 thermal receipts across five locales, each with a photo-degraded twin, pixel-exact word boxes and KIE fields 32k docs · 4.3 GB · 1k+ downloads
lazy-cat Claude Code skills that stop the agent from over-engineering: think before choosing an approach, ask before adding what nobody requested 18.6× fewer tokens across 17 tasks · 50★
OpalZero Multi-agent engine in Rust: a plain-English intent in, typed and verified state out over SSE Planner → Dispatcher → Governor

On Hugging Face: 10 models · 4 datasets · 10k+ downloads — huggingface.co/albertobarnabo

Technical Expertise

Stack & Infrastructure
Languages Python Rust Java TypeScript C
AI & ML Infrastructure PyTorch Hugging Face LangChain FastAPI Weights & Biases
Cloud & Data AWS Azure Docker Terraform Kubernetes VectorDB PostgreSQL

Research

  • Large Language Models for Fact-Checking over Tabular Data — Master's thesis, Xi'an Jiaotong University & Politecnico di Milano (2024). How LLMs verify claims against tables (TabFact, FEVEROUS) and how prompting strategy changes accuracy. Summary (PDF) · Dataset · Code

Academic Background

  • M.Sc. Computer Science Engineering — Politecnico di Milano (105/110)
  • M.Sc. Computer Science Engineering — Xi'an Jiaotong University (Double Degree)
  • B.Sc. Computer Science Engineering — Politecnico di Milano

Pinned Loading

  1. lazy-cat lazy-cat Public

    Claude Code skills for developers who code like cats — never more effort than the problem requires.

    JavaScript 51 1

  2. Fact-Checking-Pipeline Fact-Checking-Pipeline Public

    Fact Checking pipeline for the FEVEROUS dataset implemented using LangChain

    Jupyter Notebook

  3. fiduciary fiduciary Public

    A senior personal financial advisor: Qwen3-4B fine-tuned with LoRA + live market/news tools, fully local on Apple Silicon (MLX).

    Python 1

  4. opal-zero opal-zero Public

    Open-source multi-agent AI engine written in Rust. Turns any intent into coordinated research, analysis, and synthesis.

    7

  5. bibliotech bibliotech Public

    Claude's personal library, organised as shelves. Built to survive across sessions so research does not have to be redone.

    Python 2

  6. product-search-embeddings product-search-embeddings Public

    Self-hosted AI product search for e-commerce — embeddings + reranker trained on 427k real Amazon relevance judgments. Runs on CPU. No per-query fees.

    Python 2