Skip to content
View lucalullo's full-sized avatar

Block or report lucalullo

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. 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
lucalullo/README.md

Luca Lullo

Data Scientist & Machine Learning Engineer

Python · SQL · Tabular Machine Learning · LLMs · AI Agents

I develop open-source Python tools and machine learning projects, focusing on tabular learning, feature engineering, model evaluation, and reproducible experimentation.

My work ranges from experimental algorithms and reusable Python libraries to applied data analysis, language models, AI agents, and machine learning implementations from first principles.

I prioritize automated testing, clear documentation, reproducibility, and transparent reporting of results and limitations.

Open-source Python libraries

Three of my machine learning libraries are available on PyPI.

CodAdapt

GitHub · PyPI

A Python library for tabular classification and regression based on adaptive coded memory and shared multi-resolution encoding.

  • Binary and multiclass classification, and regression
  • scikit-learn-style estimator API
  • Automatic handling of mixed tabular data
  • Experimental modeling approach with a documented API

Install: pip install codadapt

MorphTree

GitHub · PyPI

An experimental tree-based machine learning library exploring structural evolution of learning algorithms for tabular data.

  • Binary and multiclass classification, and regression
  • Structural mutations guided by model evaluation
  • Automatic handling of numerical, categorical, and missing data
  • Python and native C components with a scikit-learn-style API

Install: pip install morphtree

NeuroTabular

GitHub · PyPI

A PyTorch-based library for tabular classification and regression, designed to simplify neural network workflows on structured data.

  • Binary and multiclass classification, and regression
  • Automatic numerical and categorical preprocessing
  • scikit-learn-compatible API
  • Model evaluation, testing, and reproducible workflows

Install: pip install neurotabular

Kaggle

Expert in Datasets and Notebooks

  • Datasets: #21 of 12,348 (15 medals)
  • Notebooks: #173 of 57,482 (35 medals)

I publish datasets and notebooks focused on data quality, exploratory analysis, feature engineering, and predictive modeling.

Kaggle Profile

Selected projects

A selection of projects spanning machine learning, data analysis, software development, language models, AI agents, and implementations from first principles.

Project Description
Predicting Electric Vehicle Purchases Custom gradient boosting implementation from scratch using NumPy and pandas, achieving 0.93827 private ROC-AUC in a Kaggle competition.
BurOCRazia Privacy-focused offline application for recognizing Italian administrative documents using OCR, deterministic matching, and a traceable SQLite catalog.
Home Credit Default Risk Credit default prediction using multi-table feature engineering, XGBoost, Optuna optimization, and SHAP-based interpretation.
Italian Justice System Analysis Exploratory analysis of Italian judicial workload from 2003 to 2024, examining civil and criminal proceedings, long-term trends, and population-adjusted indicators.
Building Agentic AutoML Experimental AutoML system combining automated preprocessing, model selection, cross-validation, and rule-based experiment planning.
Building AI Agent Step-by-step implementation of AI agent components in Python, including local LLM-based tool routing, argument parsing, memory, and multi-step execution.
Customer Support Agent Educational customer support agent built with Google ADK and Gemini, featuring automated request classification and human-in-the-loop escalation.
Building LLM Language models implemented from first principles, from statistical approaches to a small decoder-only Transformer with training, evaluation, and instruction tuning.
Building LLM Technologies Ten educational notebooks implementing core LLM mechanisms, including causal masking, positional encoding, multi-head attention, Transformer blocks, and token sampling.
Building Neural Networks Neural networks implemented from scratch using NumPy, covering backpropagation, mini-batch training, and deep networks evaluated on Fashion-MNIST.
Building Decision Trees Decision trees and gradient boosting implemented from first principles, including numerical and categorical splits, recursive growth, and missing-value handling.

Technical skills

Languages: Python, SQL

Data analysis: pandas, NumPy, Jupyter

Machine learning: scikit-learn, XGBoost, LightGBM, Optuna, SHAP

Deep learning: PyTorch, TensorFlow

LLMs & AI Agents: Transformers, tokenization, attention mechanisms, language modeling, tool calling, agent orchestration

Development: Git, GitHub, automated testing, Python packaging, reproducible experiments

Cloud: Google Cloud

Courses & certificates

Selected training programs:

Connect

LinkedIn · Kaggle

Pinned Loading

  1. codadapt codadapt Public

    Experimental tabular machine-learning algorithm based on adaptive coded memory and shared multi-resolution encoding.

    Python

  2. morph-tree morph-tree Public

    Native tree-based machine learning with structural algorithm evolution for binary classification, multiclass classification, and regression.

    Python

  3. neuro-tabular neuro-tabular Public

    Neural networks for tabular data with automatic preprocessing and a scikit-learn-style API.

    Python

  4. predicting-electric-vehicle-purchases predicting-electric-vehicle-purchases Public

    Custom gradient boosting implementation from scratch for electric vehicle purchase prediction. Kaggle competition private score: 0.93827 ROC-AUC.

    Jupyter Notebook

  5. burocrazia burocrazia Public

    Applicazione locale e deterministica per riconoscere con prudenza documenti amministrativi italiani, con OCR, catalogo tracciabile e funzionamento offline.

    Python 3

  6. building-llm building-llm Public

    A step-by-step educational journey from a character-level statistical language model to a small decoder-only Transformer.

    Jupyter Notebook 2 1