Riccardo Riva

A text assistant to help users to surf the public payment system website in Apulia and Veneto. Built with Google DialogFlow and a fulfillment server use APIs and Protobuf library to create responses user based. DialogFlow classify the message and based on the intent recognized the fulfillment server create a response.

Working on a merge of six databases to a datawarehouse. for Apulia region in Italy with thousands of tables and millions of records. The team is made of 10 swe coming from different working areas and our goal is to let it flow from an old 90s DB to a state of the art solution. Building pipelines to correct and cast fields with the best practices and removing old patches that slow down queries.

An Amazon Alexa Skill for vocal suggestions on places to visit and events to attend in the Apulia region. User can ask for any specific town in Apulia and an action will retrive information. We fine tuned a SpaCy model to do the NER (Name Entity Recognition) for every place in the region.

Student of physics at the first university of Rome. Laboratory classes had a great impact on me thanks to the teamwork, analysis of data, retrieval them, building error propagation theories. My thermodynamics teacher was a collegue of Barry Barish (Nobel Prize winner), they also had a class together.

Learn the basis of Machine Learning, Deep Learning, Data Analysis, Data visuaization, RAG, Fine Tuning, Data preparation, how to choose and build a model from the ground (or from white papers) for Computer Vision, Transformers, NLP, and much more.

arxiv.org/abs/2104.14045 In this paper we recreate, and improve, the binary classification method for particles proposed in Roe et al. (2005) paper "Boosted decision trees as an alternative to artificial neural networks for particle identification". Such particles are tau neutrinos, which we will refer to as background, and electronic neutrinos: the signal we are interested in. In the original paper the preferred algorithm is a Boosted decision tree. This is due to its low effort tuning and good overall performance at the time. Our choice for implementation is a deep neural network, faster and more promising in performance. We will show how, using modern techniques, we are able to improve on the original result, both in accuracy and in training time by about 1% (from 94.5% to 95.2%).