Arxelos deploys research-grade AI models at the intersection of neuroscience and deep learning. To replicate the astonishing capabilities of the human brain, we must first understand how it works. Each model is a working artifact — built to demonstrate, not just describe.
Each model sits at the intersection of AI capability and biological insight — deployed, demo-ready, and built to withstand technical scrutiny.
4-class classification on 7,023 MRI scans. 96.2% test accuracy. A deep CNN deployed with FastAPI, containerized with Docker, and running on Azure.
Ablate specific layers and channels in VGG-16 or ResNet-50. Measure whether prediction deficits mirror biological visual deficits — prosopagnosia, akinetopsia, achromatopsia.
Ingest PubMed abstracts, build a retrieval pipeline, generate domain-specific answers with source citations. Medical RAG with cited sources — not another generic chatbot.
An 8-week independent study mapping biological neural systems to deep learning architectures. Every lab produces a deployable artifact.
Neuroanatomy, Hodgkin–Huxley neuron models, Leaky Integrate-and-Fire simulations, and mapping the human visual stream against CNN feature hierarchies.
Sparse coding, synaptic plasticity, STDP learning rules, Hopfield networks, place cell representations, and RL agents with biologically-inspired state encodings.
Language processing parallels, mechanistic interpretability, circuit analysis, and a capstone BCI neural signal decoder using EEG motor imagery data.
The theoretical backbone — primary literature driving each phase of the study.
Thoughts on neuroscience, deep learning, and the space between — written as I build.
Every neural network you've ever trained learned through backpropagation. Every neuron in your brain learned without it. What AI lost by abandoning biology — and why the field is circling back.
Google's GraphCast can generate a 10-day forecast in under a minute. But when researchers tested AI models on Category 5 hurricanes, they failed catastrophically. What this means for AI in safety-critical domains.
What happens when you break a neural network the same way a stroke breaks a brain? Exploring the parallels between CNN ablation studies and biological visual deficits.