I specialize in Deep Learning, Graph Neural Networks (GNNs), and Mechanistic Interpretability, building systems that scale from fundamental physics research to conversational AI.

• Mechanistic Interpretability: Diagnosing structural hierarchy in transformer models. Using Sparse Autoencoders (SAEs) to isolate depth-tracking features and verify causality within neural activations via forward-pass hooks.
• High-Energy Physics via GNNs: Developing progressive non-local Graph Neural Network pipelines for particle jet classification, pushing performance boundaries while optimizing for memory efficiency with O(N) scatter ops.
• Agentic Long-Term Memory: Architecting multi-tier, asynchronous retrieval systems to bound context window growth and slash latency across thousand-turn LLM conversations.
• Adversarial Audits & Fairness: Quantifying demographic disparities in predictive health metrics and auditing complex-valued neural beamformers against adversarial (FGSM) attacks to expose phase-blind vulnerabilities.

• Languages: Python (Advanced), C++ (Basics), Go (Basics).
• Deep Learning & ML: PyTorch, TensorFlow, Scikit-learn, XGBoost, GNNs, LLMs & RAG, Time Series.
• Libraries & Ecosystem: Pandas, NumPy, Matplotlib, NetworkX, TransformerLens, SAELens.
• Infrastructure & Dev Tools: FastAPI, Qdrant (Vector DB), Git/GitHub, Jupyter, Groq LPU, REST APIs.

• Scientific Ecosystem Contributor: Actively engineering features for global open-source libraries used in neuroinformatics, ecology, and astronomy.
• Algorithmic Implementation: Built mathematical functions for spatial plotting, complex kinematics (turning angles, path sinuosity), and absolute feature selection estimators.
• System Reliability: Resolved critical missing-data semantics, patched downstream failures via early input validation, and stabilized CLI error handling in large-scale codebases.
• Competitive ML: Top-tier finalist in a global hackathon for long-term conversational memory, achieving massive speedups over sequential retrieval systems.

Deep Learning Enthusiast | Open-Source Advocate | Minimalist Programmer
I build for interpretability, mathematical rigor, and real-world impact. Always exploring the latent space, squashing bugs, and reimagining what's possible with a GPU.

Let's connect, collaborate, or just nerd out about loss functions.