RcppDeepState

RcppDeepState is an R library to fuzz test compiled code in Rcpp packages. The package extends the DeepState framework to fully support Rcpp based packages. RcppDeepState allows to fuzz test R library’s C++ code in order to find more subtle bugs like memory leaks or even more general memory errors.

RcppDeepState-action

I am the author of RcppDeepState-action, a GitHub Action which enables users who develop R packages using Rcpp to find memory leaks in their code that is hosted on GitHub.

R-integration

I am the author of R-integration, a Node.js library that allows to execute arbitrary R commands or scripts directly from the Node.js environment. This integration works on Windows and GNU/Linux based systems.

Open source contributions

DeepState

Author of the entire Windows implementation of the framework. Before this, DeepState was restricted to POSIX-compliant GNU/Linux or macOS operating systems. Merged into DeepState in 2023

NASA FPrime

Resolved potential overflow issue arising from overlapping strings in memory during copying operations. Merged into FPrime in 2021
Improved code quality and conducted reviews of string manipulation framework utilities. Merged into FPrime in 2021

Italian digital public services

Led team effort to devise a Docker-based system for aggregating and monitoring vast data sets.
Integrated 4 services with Docker-compose reducing to few seconds the time needed to aggregate and visualize data. Merged into daf-recipes in 2017

The R Project for Statistical Computing

I incorporated RcppDeepState into packages that can be accessed on both CRAN and GitHub. This enables the identification and reporting of memory-related problems within these packages. Through GitHub Actions’ continuous integration system, developers can effortlessly assess their R packages for memory-related issues, eliminating the need to manually configure fuzzers and other memory detection utilities.

This is a list of the packages that RcppDeepState-action automatically checks:

Coursework

A Comparative Study of Genetic-Based Approaches for Enhanced Hourly Temperature Predictions

A comparative study of two genetic-based approaches for enhancing hourly temperature predictions. The first approach employs a LSTM optimized through Genetic Algorithms for hyperparameter tuning. The second approach utilizes Genetic Programming to create an interpretable model. The study showcases the strengths and weaknesses of both approaches, with the LSTM model outperforming the Genetic Programming model in terms of Mean Absolute Error, while the Genetic Programming model provides an interpretable model, which might be preferred in scenarios where efficiency is important.

CLIP Region Proposal Network

A deep learning framework, ClipRPN, to tackle visual grounding, a challenge linking language and perception by grounding linguistic symbols in visuals. Built upon the CLIP model from OpenAI, the solution converts images and textual descriptions into feature maps and embeddings, which are then fused and processed by a Region Proposal Network to predict bounding boxes in images based on the textual information. Implemented and evaluated the model using the RefCOCOg dataset, showcasing the model’s ability to generate accurate bounding boxes around referred objects in images with an overall intersection over union(IOU) of 41.19%.

Hybrid recommendation system

Designed a cutting-edge query recommendation system focused on revolutionizing query recommendations within a Database Management System. Leveraged a unique combination of Collaborative Filtering and Locality Sensitive Hashing (LSH) to optimize recommendation accuracy and time efficiency. Showcased the seamless integration of LSH for improved item similarity searches and introduced a content-based strategy to develop a hybrid recommendation system, resulting in a improvement in query recommendation accuracy.

Distributed Key-Value Store in Akka

Inspired by Amazon Dynamo, this project introduces a DHT-based peer-to-peer key-value storage service implemented using Akka actors. It provides a simple user interface to upload/request data and issue management commands, emphasizing key principles of data partitioning, replication, and dynamic network management.

Standalone Stellar Blockchain

A ready-to-use setup of a standalone Stellar blockchain based on Docker that I implemented as part of my thesis study. Implementation of a Stellar validator node that runs on a fully private blockchain.

LR(0) automata generator

Parsing automa generator written in C++ created as final project for the ”Formal Languages and Compilers” course at University of Trento.