𧩠I make things, break things, and occasionally figure out why they broke.
ποΈ I've built ML systems at Siemens, wrangled startups at AWS, and now lead AI at Critical Software.
π± I'm on a mission to bring ML to highβstakes domains and build AI that solves real problems for real people.
π¦ΉπΌββοΈ As you're reading this, I'm probably out there in the world doing something mildly heroic, questionably wise, or both.
π« Want to say HELO? Find me on LinkedIn (fast), send me an email (please don't!) or use a carrier pigeon.
Small models for health and life sciences data. Research aids, not for clinical use.
| Item | Type | Tags | Description |
|---|---|---|---|
| mosquito-mtrcnn-dg | Model | audio-classification bioacoustics mosquito domain-generalization |
A compact MTRCNN classifier that names the mosquito species in a flight-tone recording, trained for cross-domain generalization on the BioDCASE 2026 Task 5 development data. It... |
| mosquito-species-demo | Space | gradio |
Name the mosquito species in a flight-tone recording |
Tools and datasets for looking at how models portray people.
| Item | Type | Tags | Description |
|---|---|---|---|
| timbre-explorer | Space | gradio |
Hear how occupation prompts change TTS voices |
Small models pre-trained from scratch on a budget.
| Item | Type | Tags | Description |
|---|---|---|---|
| eca | Model | pytorch safetensors portuguese literature from-scratch |
Small GPT pre-trained from scratch on the collected works of the Portuguese novelist Eça de Queirós, from the jgalego/eca-queiros dataset. Byte-level BPE tokenizer trained on... |
Models for underrepresented languages and dialects.
| Item | Type | Tags | Description |
|---|---|---|---|
| amalia-desenrasca-9b | Model | transformers safetensors llama text-generation tool-calling |
amalia-llm/AMALIA-9B-0626-SFT fine-tuned to call tools, with prompts in European Portuguese (pt-PT). Desenrascanço is the Portuguese art of improvising your way out of a mess. |
| desenrasca-demo | Space | gradio |
AMALIA calls tools in European Portuguese |
Small models past the standard transformer.
| Item | Type | Tags | Description |
|---|---|---|---|
| trm-sudoku-extreme | Model | safetensors sudoku recursive-reasoning trm arxiv:2510.04871 |
Tiny Recursive Model (TRM) trained from scratch to solve Sudoku-Extreme puzzles. One small network recursively refines a latent state and an answer, from Less is More: Recursive... |
| breakout-world-model | Model | pytorch safetensors world-model diffusion atari |
A diffusion world model of Atari Breakout, trained from scratch. Given the last 4 frames and actions, a small U-Net denoises the next 64Γ64 frame; fed its own frames, it... |
| buridan-2b | Model | peft safetensors decision-model pointer-head calibration |
A decision model: it picks one of the options it is given and says how sure it is, in one forward pass, without generating text. A replication of Strands Decider on... |
| doom-world-model | Model | pytorch safetensors world-model diffusion doom |
A diffusion world model of Doom, trained from scratch on the first level of Freedoom (E1M1) in ViZDoom. Given the last 4 frames and actions, a small U-Net denoises the next... |
| buridan-demo | Space | gradio |
Pick one option from a text and say how sure |
| space-invaders-world-model | Model | pytorch safetensors world-model diffusion atari |
A diffusion world model of Atari Space Invaders, trained from scratch. Given the last 4 frames and actions, a small U-Net denoises the next 64Γ64 frame; fed its own frames, it... |
| ms-pacman-world-model | Model | pytorch safetensors world-model diffusion atari |
A diffusion world model of Atari Ms. Pac-Man, trained from scratch. Given the last 4 frames and actions, a small U-Net denoises the next 64Γ64 frame; fed its own frames, it... |
| trm-sudoku-demo | Space | gradio |
Watch a 5M-parameter model solve Sudoku step by step |
| fly-whisperer | Space | static |
Sneak up on a fruit fly, then teach its brain a smell |
Small models whose output is checked by a compiler, prover or model checker.
| Item | Type | Tags | Description |
|---|---|---|---|
| ada-coder-qwen2.5-1.5b | Model | transformers safetensors qwen2 text-generation ada |
Qwen/Qwen2.5-Coder-1.5B-Instruct fine-tuned with LoRA to write Ada 2022 and SPARK code from a task description, spec or signature. |
| ada-coder-qwen2.5-7b | Model | transformers safetensors qwen2 text-generation ada |
Qwen/Qwen2.5-Coder-7B-Instruct fine-tuned with LoRA to write Ada 2022 and SPARK code from a task description, spec or signature. |
Small models for aviation and safety engineering. Research aids, not certified.
| Item | Type | Tags | Description |
|---|---|---|---|
| notam-subject-qwen3.5-0.8b | Model | transformers safetensors qwen3_5_text text-generation aviation |
Qwen/Qwen3.5-0.8B fine-tuned with LoRA to name the subject of a NOTAM: one of the 13 classes in DEEL-AI/NOTAM. |
| reqlint-smollm3-3b | Model | transformers safetensors smollm3 text-generation requirements-engineering |
HuggingFaceTB/SmolLM3-3B fine-tuned with LoRA to check a requirement for common writing defects and rewrite it in EARS form. Where the rewrite needs information the original... |
| c172-flight-dynamics | Model | pytorch safetensors aviation world-model flight-dynamics |
A small world model of a Cessna 172 in flight. Given the last 4 aircraft states and stick commands, it predicts the state 0.1 s later; fed its own predictions, it flies the... |
| ppo-turn-heading-cessna172p | Model | stable-baselines3 aviation jsbsim gymnasium ppo |
A PPO agent that flies a Cessna 172P in the JSBSim flight dynamics model. It starts on a random heading at 5,000 ft and has to turn onto a random target heading and hold... |
| notam-subject-demo | Space | gradio |
Classify NOTAM text into its main aviation subject |
| turn-heading-demo | Space | gradio |
Pick a heading and watch a PPO agent turn a Cessna |
Models and Spaces that have not been added to a collection yet.
| Item | Type | Tags | Description |
|---|---|---|---|
| knot2vec | Model | pytorch safetensors knot-theory embeddings contrastive-learning |
Embeddings for knot diagrams. Two diagrams of the same knot, however tangled, should land close together; diagrams of different knots should land apart. Part of Weird2Vec,... |
| knot2vec-demo | Space | gradio |
Tangle a knot diagram and see if Knot2Vec can still name it |
Older models, datasets and Spaces.
| Item | Type | Tags | Description |
|---|---|---|---|
| rl_course_vizdoom_health_gathering_supreme | Model | sample-factory tensorboard deep-reinforcement-learning reinforcement-learning |
A(n) APPO model trained on the doom_health_gathering_supreme environment. |
| llava-onevision-qwen2-0.5b-ov-hf | Model | transformers onnx safetensors llava_onevision image-text-to-text |
This repository contains code and instructions for deploying the LLaVA-OneVision multimodal model on Amazon SageMaker using the Hugging Face Inference Toolkit. |
| tokenizers-languages | Space | gradio |
Comparing LLM tokenizers in multiple languages |
π§ͺ More experiments live on Hugging Face: browse all my models and try the full collection of Spaces.
@langchain-ai/langchain β The agent engineering platform.
@ScrapeGraphAI/Scrapegraph-ai β Python scraper based on AI
@huggingface/smolagents β π€ smolagents: a barebones library for agents that think in code.
@highlightjs/highlight.js β JavaScript syntax highlighter with language auto-detection and zero dependencies.
@bendlang/bend β Bend 2: a fast language that blocks AI mistakes via proof. Install: curl -fsSL https://bend-lang.com/install.sh | sh
@SWE-agent/SWE-agent β SWE-agent takes a GitHub issue and tries to automatically fix it, using your LM of choice. It can also be employed for offensive cybersecurity or competitive coding challenges. [NeurIPS 2024]
@huggingface/text-embeddings-inference β A blazing fast inference solution for text embeddings models
@AutoCodeRoverSG/auto-code-rover β A project structure aware autonomous software engineer aiming for autonomous program improvement. Resolved 37.3% tasks (pass@1) in SWE-bench lite and 46.2% tasks (pass@1) in SWE-bench verified with each task costs less than $0.7.
@langchain-ai/langserve β LangServe π¦οΈπ
@google-deepmind/concordia β A library for generative social simulation
@JGalego/awesome-safety-critical-ai β When the stakes are high, intelligence is only half the equation - reliability is the other
@JGalego/RunNX β Fast, fearless, and fully verifiable ONNX runtime in Rust πβ‘π¦
@JGalego/eeg-bci-tutorial β EEG-based BCI tutorial based on MNE, SciKit-Learn and PyRiemann
@JGalego/RAGmap β A simple Streamlit application to visualize document chunks and queries in embedding space πΊοΈπ
@JGalego/Hack-GraphRAG β Learn how to run GraphRAG pipelines backed by Amazon Bedrock using LiteLLM proxy π
@JGalego/deploy-langserve-aws β Learn how to deploy π¦π LangChain applications with π¦οΈπ LangServe in minutes on Amazon ECS and AWS Fargate using AWS Copilot.
