Skip to content
View HenryNdubuaku's full-sized avatar
  • Cactus Compute
  • London
  • 11:37 (UTC +01:00)

Block or report HenryNdubuaku

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
HenryNdubuaku/README.md

Henry Ndubuaku

LinkedIn Twitter Email Spotify

I could train a 1B-A200m model on an iPhone 17 Pro at ~650 tokens/sec. It will take 360 days on 20B tokens of training data and use 156KW of electricity which cost $51.

The phone will fry of course, so I wrote algorithms to run inference on your phone rather. We named it after a plant that survives in resource-constrained environments, the Cactus.

cactus can run similar model on your Grandma’s Pixel 6a at 80 tokens/second while only draining 10% battery per hour of continuous inference and using 250MB RAM only.

I then built an agentic foundation model for tiny smart devices, its compute graph is slim like a Needle. Needle has no FFN, just 26m params of pure attention, yet matches 10-25x bigger models. It can be a better Siri/Alexa for tiny smart devices, so we spun into a company with my friend.

We did Y Combinator, and raised from FCVC (Slack, Coinbase, GitLab, Instacart etc.), Samsung Next, Oxford's Seed Fund, 6 smaller funds like Transpose (run by Garry Tan's brother), fellow YC founders, and 62 tech CTOs/VP/Dir both via syndicate and directly at DeepMind, Palantir etc.

Cactus now powers cool products you've probably heard of...I think.

6 gifted Research Engineers have joined us:

  • Karen is an International Physics Olympiad Gold Medalist...says he's 22 (I don't believe him).
  • Satyajit joined Fei Fei Li's Lab at Stanford at 14y, I've been accused of coddling him...he's 21!
  • Justin already interned at OpenAI and published at NeurIps, this one can sell me for money...he's 19.
  • Noah previously interned at Google & Microsoft, an idea machine gun...when not crashing out...he's 21.
  • Jakub doesn't talk, just turns up with flawless execution, the company's John Wick...he's 22.
  • Parkirat is basically my boss ATP, finally learnt to speak up, probably the best engineer...just 21.

They are all smarter than I am by sheer IQ, I've just done more things haha.

Cactus really is a tiny team, building tiny AI for tiny devices :)

Publications

Parameter-Efficient Transformer Embedding via Functional Factorization (ICML 2026)
HiDRA: A Blazing Fast LM-Head Replacement (ICLR 2026)
Depth Over Specialization in Small Multimodal Transformers (ICLR 2026)
Just Enough Learning: GRPO-Guided Controllers for Hyperparameter Sweeps (ICLR 2026)
TACE: Token-Aware Chunked Encoding For Realtime Speech Models (ICLR 2026)
CLAWS: Calibration-Aware Activation Sparsity for Instruction-Tuned LLMs (ICML 2026)

Fun Facts

  • After CUDARepo, Nvidia reached out, I did 7 technical rounds, got a verbal offer, back-and-forth over YOE/pay, then I got YC.
  • Did MSc at QMUL, just to work with Prof Matt Purver (Ex-Stanford Researcher on CALO), did my project/thesis with his team.
  • Did BEng under Prof Onyema Uzoamaka (Rumoured first Nigerian CS grad from MIT), he taught computing archs off-head!
  • Biggest career miss was a PhD Studentship at Meta FAIR.

Personal Life

Profile Nigerian-British, born Jan 1996, 185cm, 83kg
Hobbies Calisthenics, UFC, chess, music
Philosophy Humanist Christian, unpolitical

Speaking at UCLA

The Cactus Jacks

Team dinner

I'm fit too

Cactus Jacks at YC

Cactus Pod at YC HQ

Music Profile

Expressive Rap
Alternative/Folk
Soul/Jazz
Oldies
Genre-Blending Urban
Dark Pop

Pinned Loading

  1. maths-cs-ai-compendium maths-cs-ai-compendium Public

    Become a cracked AI/ML researcher/engineer with this unconventional textbook covering maths, computing, and ML with intuition.

    TypeScript 7.3k 898

  2. cactus-compute/needle cactus-compute/needle Public

    14MB foundation model for tiny devices; phones, wearables, smart home, and robots.

    Python 5.8k 387

  3. cactus-compute/cactus cactus-compute/cactus Public

    Quantization, kernels, runtime and inference engine for mobiles, wearables, smart home and robots.

    C++ 5.8k 471

  4. vector vector Public

    Programming language for ML on XLA with JAX-level speed; compiles for CPU, TPUs/Nvidia GPUs, AMD GPUs, Apple GPUs, etc.

    Rust 15 6