Projects

Things that I built, at work and on the side.

  1. 2023 – 2025

    Lilli, McKinsey’s generative AI platform

    McKinsey’s firm-wide generative AI assistant, built at QuantumBlack. Consultants ask it questions and it answers from the firm’s knowledge base and expert network, with sources. I built its core.

  2. 2022

    One-shot face recognition

    A one-shot face-recognition model for J.P. Morgan. Used triplet loss and autoencoders for data augmentation, improving prediction accuracy by 16%.

  3. 2022

    Species presence prediction

    Multimodal transformers trained on 600 GB of satellite imagery to classify species. Built custom batching and memory-loading techniques to train within limited GPU memory.

  4. 2022

    Music generation with RNNs

    An LSTM-based recurrent neural network in TensorFlow that generates piano music by learning from its structure and timing.

  5. 2022

    Natural language to satellite imagery

    An application that converts natural-language prompts into satellite images using GPT-3 and retrieval-augmented generation. Third place at an OpenAI competition.

  6. 2021

    Housing-price prediction

    A ProperBird project using web-scraped German housing data and XGBoost, linear regression, and CatBoost to predict prices.