amaarmc.org · optimizer research · code intelligence · agent coordination
I build AI-native systems at the intersection of neuromorphic ML research, personal agents, and developer infrastructure.
Current anchor work:
- Princeton / MedARC MindEye: neuromorphic computing research for real-time neural decoding with fMRI, DNNs, and transformer-based vision models.
- null0.ai: a sandboxed AI clone that connects to your platforms, learns your context, and operates on your behalf.
- FastMuon: local optimizer research testing whether reduced-depth Newton-Schulz Muon variants preserve sample-efficiency gains.
Princeton · MedARC MindEye · fMRI → vision models
Researching real-time visual perception decoding by aligning brain activity with deep vision models. Current threads include TR-by-TR inference, retrieval/reconstruction metrics, fast voxelwise beta estimation, and subject-specific model adaptation.
Princeton research MedARC MindEye neuromorphic ML fMRI neural decoding
sandboxed AI clone · personal context · platform actions
Building a macOS beta that consolidates messages, relationships, and workflows into one hub. The core idea is a personal agent that understands your context and acts across the platforms you already use.
FastMuon · Newton-Schulz depth · PyTorch sweeps
Built a local research harness for Muon-family optimizers. First signal: NS3/NS4 reduced-depth Muon preserved much of Muon's small-LLM validation-loss improvement while using fewer Newton-Schulz iterations.
fastmuon-local-research NS3 NS4 tested TinyGPT sweeps
MCP · codebase context · coordination protocols
Working on local-first code intelligence and coordination layers for coding agents: dependency graphs, symbol-aware context, declared intents, locks, and conflict detection before agents edit the same code.
n8n · operations workflows · model evaluation
Designed production automations for operations, finance, routing, and AI-assisted decision support. Also built evaluation workflows for model behavior, prompt reliability, and agentic workflow consistency.
n8n automation model evals production workflows
Cotton · wardrobe AI · outfit generation
Founded Cotton, an AI closet product for clothing detection, outfit generation, and personal styling. The work spans product vision, AI integration, growth, fundraising, and team execution.
Founder Cotton computer vision clothing detection product strategy
Python PyTorch TensorFlow NumPy OpenAI TypeScript JavaScript React Next.js Svelte Vue Supabase PostgreSQL Rust C++ Git GitHub Docker
Operating principle: research to prototype to measure to ship
