Henry Ndubuaku

Researcher working across NeuroAI, neural signal modelling and efficient neural architectures. Co-founder and CTO of Cactus Compute (Y Combinator). London and San Francisco.

X GitHub LinkedIn
  1. Cactus Compute (Y Combinator), Co-founder and CTO2025–present
    San Francisco / London
    • Eight-person research team on efficient foundation-model architectures.
    • Multimodal learning, compression, structured memory, fast inference.
    • Leads the team behind Cactus Whistle, Cactus Quants and Cactus Hybrid.
    • Six 2026 papers at ICLR, ICML and NeurIPS NeuReps, listed under Works.
  2. Deep Render, AI Research Engineer2024–2025
    London
    • Real-time neural video models with compression, vision and diffusion.
    • Hardware-aware inference and CPU, GPU and NPU kernels.
    • Multimodal speech and vision training infrastructure.
  3. Wisdm AI, Machine Learning Software Engineer2021–2024
    London
    • Visual-perception models from satellite imagery and LiDAR.
    • Segmentation, 3D reconstruction and neural radiance fields.
    • GPU-accelerated inference and signal-processing pipelines.
  4. Queen Mary University of London, MSc Artificial Intelligence2020–2021
    • Under Prof Matthew Purver.
    Modules
    AI in Games (Search, Evolution & RL) · Data Mining · Machine Learning · Computational Linguistics · NLP & Neural Networks · Deep Learning & Computer Vision · Cognitive Robotics · Advanced Robotics
    Research
    Improving Creativity In Humour Generation With Conditional Stochastic Latent-Sampling Encoders
    Comparative Analysis of SOTA Vision Architectures (VGG, GoogLeNet, ResNet & Vision Transformers)
    Open source
    Maths, CS & AI Compendium · Google Developer Expert (JAX) · NanoDL model training library (author) · Vector programming language (author) · Halo on-device quantized diffusion library (author)
  5. Google Africa Developer Scholarship Program2018–2019
    • Large-scale distributed systems design.
  6. National Youth Service Corps, Software Engineer2017–2018
    • Posted to data science and ML after military camp.
  7. Madonna University, BEng Electrical Engineering2012–2016
    Maths
    Maths I–II · Linear Algebra · Calculus · Stats & Probability I–II · Computational Maths · Engineering Maths I–IV
    Physics
    General Physics I–III · Waves & Optic Physics · Electromagnetism · Applied Mechanics
    Electronics
    Circuit Theory I–II · Control Systems I–II · Physical Electronics · Analog Electronics · Digital Signal Processing
    Computing
    Digital Electronics · Microprocessors · Programming I–II · Assembly & Machine Language

Publications

  1. Cross-Context Neural Transport2026
    NeurIPS NeuReps Proceedings, PMLR
  2. Parameter-Efficient Transformer Embedding via Functional Factorization2026
    ICML
  3. HiDRA: A Blazing Fast LM-Head Replacement2026
    ICLR
  4. Depth Over Specialization in Small Multimodal Transformers2026
    ICLR
  5. Token-Aware Chunked Encoding for Realtime Speech Models2026
    ICLR
  6. GRPO-Guided Controllers for Hyperparameter Sweeps2026
    ICLR
  7. A Controlled Study of Attention-Only Transformers2026
    arXiv:2607.18363
  8. Modelling The Brain's Response To Natural Scenes In The Bottleneck Space2023
    bioRxiv
  9. Improving Creativity In Humour Generation With Stochastic Conditional Latent-Sampling Encoders2021
    With R. Shekhar and M. Purver
  10. NeuroAI write-ups
    Two notes on Overleaf: part one and part two.

Projects

  1. Maths, CS & AI Compendium
    Open-source notes written gradually over seven years.
  2. Cactus Needle
    Compact model family with adaptive depth.
  3. Cactus Kernels
    Inference kernels for Cactus, most of them written by hand.
  4. Hadamard MLP
    Cactus Compute blog.
  5. Intelligence Ladders
    Cactus Compute blog.
  6. Simple Attention Networks
    Cactus Compute blog.
  7. Vector
    Lightweight language for machine learning on the XLA compiler.
  8. NanoDL
    JAX library for training sub-4B foundation models for the edge.
  9. R² Index
    Global ranking of AI researchers, papers, startups and investors by measurable impact.
  10. Cactus Whistle, Cactus Quants and Cactus Hybrid
    Team projects at Cactus Compute. I lead the research, engineering and team behind them.
  1. Why I built R² Index12 Oct 2026
    A ranking of AI researchers by what they have actually produced, and how it rediscovered AI lore on its own.