Nathan Baune
Platform & Systems Engineer
I build production tools, ML systems, and complex software platforms — from real-time EEG classification and neuroimaging pipelines to a GPU-accelerated simulation engine with its own visual design environment, multi-language CLI, and LLM-powered natural language interface.
The thread connecting my work: studying how information processing architectures produce emergent behavior—from neural circuits to ML models to simulated ecosystems.
For the past decade I've built production systems in neuroscience and health tech:
- Research tools and systems shipped to 9 labs across 3 institutions.
- 2 startups: co-founder (PlatformSTL LLC), founder & principal engineer (Gothic Grandma LLC).
- MR.Flow: GUI orchestrating neuroimaging pipelines with parallel execution optimized to user hardware. Setup and deployment reduced from weeks to ~45 minutes.
- Epoche: ML workbench for neurophysiology—feature extraction, grid search, ensemble optimization, and interpretability tools for mechanistic hypothesis generation and model/ensemble deployment.
- Real-time ML classification of brain state for closed-loop experiment control, therapy onset, and neurostimulation. <40ms end-to-end latency.
- Proprio (PlatformSTL, St. Louis): STTR-funded wearable activity classification platform (hardware → cloud → ML pipeline) for monitoring patient care and long-term outcomes.
- Countless rapid-turnaround prototypes and hours of engineering, scientific, and statistical consulting.
Technologies
Current Interests
I'm actively exploring the intersection of AI and complex systems:
- Agentic AI architectures — how multi-model orchestration, tool use, and planning emerge in LLM-based systems
- Interpretability and alignment — making AI behavior legible and predictable, from SHAP/LIME to mechanistic interpretability
- Simulation as testbed — using MUSE's transparent, deterministic systems to study how architectural choices shape emergent behavior
- Human-AI collaboration — designing interfaces where AI augments rather than replaces human judgment (see: BABEL, Epoche)
Read more: How this question connects everything I've built →
Science and Engineering Spotlight
GG.Flow
2025–2026 · AlphaVisual pipeline platform for multimodal scientific research. 148 nodes across 9 domains — EEG, MRI, ML, stats, viz — with language-agnostic SDK, content-addressable caching, and auto-generation from Python, R, CLI, and YAML.
MR.Flow
2024–PresentPipeline orchestration system for MRI processing. Worker coordination, task queuing, dependency management, and failure recovery for multi-stage analysis workflows across 150+ datasets.
Epoche
2023–PresentML training pipeline for neural signal classification. Feature extraction, grid search, ensemble optimization, SHAP/LIME interpretability, and C++ serialization for real-time inference deployment.
BIDS-SQL
2026–PresentSQL schemas and dual-language libraries (Python + MATLAB) for queryable BIDS neuroimaging databases. Provenance tracking, QC metrics, EEGLAB plugin. Powers MR.Flow and Pythia data pipelines.
Proprio
2017–2021STTR-funded wearable ML platform for stroke rehab. IMU → AWS → patient-specific classifiers → clinician dashboards. Peer-reviewed publication, outperformed prior literature by 20%.
Research Portfolio
2015–2025A decade of research tools and systems across 9 labs: real-time closed-loop ML, robotic manipulandum control, VR experimental platforms, multi-device hardware synchronization, and data pipelines.
Precision Neural Engineering Lab, Emory University Medical
Neural Plasticity Research Lab, Emory University Medical
Neuromechanics Lab, Emory University Medical
Gothic Grandma LLC
2025–PresentEmergent world simulation engine (100k+ entities) with biological/psychological systems, plus the tooling ecosystem to design, debug, and experience it. Visual system designs compile directly to GPU kernels—no hand-written simulation code.
Explore the full ecosystem → to get oriented, or click on a tool below to fast-track to one that catches your eye.
Font (MUSE Sim)
GPU Simulation Engine
C++20-native simulation engine. Millions of entities, each running validated biological subsystems as neural proxy nets. GPU-accelerated, deterministic.
Babel (MUSE AI)
Semantic Interface
Multi-agent LLM pipeline executor. Model-tier resolution, targeted context injection, and parallel execution across all MUSE tools.
Glyph (MUSE Workbench)
Unified Internal Workbench
Unified Electron workbench: kernel engineering, training orchestration, AI dev, content authoring. Role-based layouts, shared Go backend.
Recent Writing
What I Learned from Years Working with "Black Boxes"
Build for inspection. Make state observable. Use AI for what it excels at.
Why ML Interpretability Needs Storytellers
Epoche: making ML legible to domain experts, not just engineers.
The Divide Between Deterministic Simulation & Black-Box AI
Why MUSE uses LLMs for translation, not generation.
From Research Pipelines to Production
Lessons learned from years operating across different engineering worlds.