Peter Graham
I tend to see the world in systems.
I'm most at home when I'm making sense of concepts that haven't quite been understood yet. I've always been drawn to ideas that are still forming - exploring new frameworks, or building one from the lack thereof. The interesting work usually lives at the seams: where ML meets neuroscience, hardware meets software, research meets product.
Most of what shows up here lands here after a long, quiet runway.
What I'm shipping, in code.
cortex-engine
Modular ML infrastructure for transformer-based neural decoders - end-to-end data pipelines, training orchestration, and a configurable serving layer built for reproducible experimentation.
visit →asclepius-research-labs
Agentic, multimodal hybrid RAG over PubMed: a research agent coordinating dense retrieval, knowledge-graph traversal, and VLM-based figure parsing for mechanism mapping, target risk, and hypothesis generation.
visit →neural-representation-explorer
Simulates population activity across 60 neurons and projects the high-dimensional response space into interpretable latent embeddings via PCA and UMAP, with decoding validation and a full CI pipeline on GitHub Actions.
visit →Writing as a way of thinking.
-
January 20, 2026
→The Infrastructure ParadoxOn why the tools we build to scale ourselves are also the ones that make us legible.
-
November 18, 2025
→The Art of Being Forward DeployedWhat FDE work taught me about the gap between a model and a deployed thing.
-
October 28, 2025
→What It Means to Feel AliveA short essay on attention, friction, and choosing the harder room.