Pitch Day – Pi School of AI Session 11

The Pitch Day event of Pi School of AI is an exciting AI event where our AI fellows present the challenges and solutions they have worked on for 8 weeks.

Our Pi School of AI Fellows pitched these challenges:

Challenge 1 - Information extraction from academic documents

  • Challenge 1

    Information extraction from academic documents
    In the “Information extraction from academic documents” challenge, we worked on partially automating the process of extracting structured information from document scans and/or pictures, the ultimate goal being to help human operators digitize information quicker. Rather than relying on Optical Character Recognition (OCR) tools, we built an end-to-end Deep Learning system that is able to extract textual information as well as structured information like the contents of a table.

  • Xinyi Chen


  • Maahin Rathinagiriswaran


Challenge 2 - Tagging incoming clients

  • Challenge 2

    Tagging incoming clients.
    Orrick is a global law firm focused on the tech & innovation, energy & infrastructure and finance
    sectors. They partnered with Pi School on a machine learning tool that draws on market and
    proprietary data to uncover sector and company insights that can inform client solutions.
    Fellows Pinku Deb Nath, Menan Velayuthan and Maria Natalia Herrera developed a Natural
    Language Processing (NLP) system that solves this problem and is easy to implement.

  • Pinku Deb Nath


  • Menan Velayuthan


  • Maria Natalia Herrera


Challenge 3 - Hierarchical Computer Vision

  • Challenge 3

    Hierarchical Computer Vision
    In the “Hierarchical Computer Vision” challenge, we demonstrate the use of our state-of-the-art computer vision model to augment the capabilities of a plant pathologist by identifying various plant diseases. With this solution, we aim to help alleviate the global food security crisis. Our model solves complex classification problems that are hierarchical in nature and can be adapted to problems in other domains as well.

  • Krzysztof Woś


  • Vijayasri Iyer


  • Dennis Rotondi


Challenge 4 - EU Sustainable Taxonomy Classifier

  • Challenge 4

    EU Sustainable Taxonomy Classifier
    We proudly present the 1st school-sponsored AIforGood challenge and the solution the fellows developed for them. The company Brink is a startup with the mission to use AI to help companies navigate their sustainability journey and accelerate the transition to a sustainable world. In the “EU Sustainable Taxonomy Classifier” challenge, we built a Natural Language Processing system that helps humans in EU Sustainable Taxonomy reporting tasks.
    A climate-specific language model has been built to upgrade existing keyword-based approaches, improve greenwashing detection, and identify truly green projects for investments.

  • Srishti Gureja


  • Vinura Dhananjaya


Pi School of AI Mentor

  • Simone Di Somma

    Managing Director of Askdata

    Simone Di Somma is the Managing Director of Askdata, a startup recently acquired by SAP that developed an AI-driven platform that allows users to answer any data-related questions with a simple search.

    Simone is an AI expert with a research focus on natural language processing, automated data analysis and human data augmentation.
    Before founding Askdata, backed by Y Combinator, Simone worked at Philip Morris and Hewlett-Packard across USA, Europe and the Middle East.

  • Filippo Galli

    Data Science PhD Student at Scuola Normale Superiore

    Filippo holds an M.Sc. in Mechatronic Engineering from Politecnico di Torino, Italy, where he graduated with a thesis on the optimal control of quadrotors carried out at the NASA Jet Propulsion Laboratory (Pasadena, California). During his studies, he took part, as Technical Student, in the Magnetic Measurement section of CERN, in Geneva, Switzerland.

    Currently, he is pursuing a PhD in Data Science at Scuola Normale Superiore in Pisa, Italy. His project aims to provide privacy guarantees to the individuals taking part in the iterative training of machine learning models, particularly in the methods of differential privacy in the context of federated learning