██████╗ █████╗ ██╗ ██╗██╗ ██████╗ ██████╗ ███╗ ██╗███████╗████████╗████████╗ ██████╗
██╔══██╗██╔══██╗██║ ██║██║ ██╔══██╗██╔═══██╗████╗ ██║██╔════╝╚══██╔══╝╚══██╔══╝██╔═══██╗
██║ ██║███████║██║ ██║██║ ██████╔╝██║ ██║██╔██╗ ██║█████╗ ██║ ██║ ██║ ██║
██║ ██║██╔══██║╚██╗ ██╔╝██║ ██╔══██╗██║ ██║██║╚██╗██║██╔══╝ ██║ ██║ ██║ ██║
██████╔╝██║ ██║ ╚████╔╝ ██║ ██████╔╝╚██████╔╝██║ ╚████║███████╗ ██║ ██║ ╚██████╔╝
╚═════╝ ╚═╝ ╚═╝ ╚═══╝ ╚═╝ ╚═════╝ ╚═════╝ ╚═╝ ╚═══╝╚══════╝ ╚═╝ ╚═╝ ╚═════╝
I'm an ML researcher and AI Engineer focused on adversarial robustness in recurrent foundation models.
My recent work — SpectralGuard — identifies and defends against Hidden State Poisoning Attacks (HiSPA) in Mamba-based State Space Models, proving that spectral radius monitoring provides a principled, real-time safety layer for SSM deployments.
I build in public: research, systems, and production AI.
My engineering philosophy is organized into a 5-layer cognitive stack, designed to separate concerns from sensory perception to immutable settlement.
graph TD
%% Nodes
L5("L5: Perception & Interface<br>(CV / Voice / Sensory)"):::percept
L4("L4: Agentic Cognition<br>(Autonomous Agents / reasoning)"):::agent
L3("L3: Semantic Memory<br>(RAG / Vector DB / Context)"):::memory
L2("L2: Transport & Compute<br>(Rust / APIs / Gateway)"):::compute
L1("L1: Settlement & Trust<br>(Blockchain / Truth Anchor)"):::trust
%% Flow
L5 --> L4
L3 --> L4
L4 --> L2
L2 --> L1
%% Styling
classDef percept fill:#0A66C2,stroke:#fff,stroke-width:2px,color:#fff
classDef agent fill:#E63946,stroke:#fff,stroke-width:2px,color:#fff
classDef memory fill:#F4A261,stroke:#fff,stroke-width:2px,color:#fff
classDef compute fill:#2A9D8F,stroke:#fff,stroke-width:2px,color:#fff
classDef trust fill:#264653,stroke:#fff,stroke-width:2px,color:#fff
|
|
|
graph TD
subgraph S1 [The Spectral Engine]
P(⚡ Complex Signal) -->|Deconstruction| D{💎 First Principles}
D -->|Architect| A[🏗️ L1-L5 Stack]
A -->|Implement| S[🚀 Scalable Production]
S -->|Telemetry| P
end
style S1 fill:#0D1117,stroke:#30363d,stroke-width:1px
style P fill:#0D1117,stroke:#A371F7,stroke-width:2px,color:#fff
style D fill:#0D1117,stroke:#00D4FF,stroke-width:2px,color:#fff
style A fill:#0D1117,stroke:#0A66C2,stroke-width:2px,color:#fff
style S fill:#0D1117,stroke:#2EA043,stroke-width:2px,color:#fff
I decompose intelligence into "fundamental frequencies" isolating core constraints (First Principles) before writing a single line of code. This ensures every system I build is grounded in physics and logic, not just hype. I don't just glue APIs; I architect vertically integrated systems.
From the low-level settlement layers (L1) and compute transport (L2) up to agentic cognition (L4) and sensory perception (L5), I engineer for reliability and scale. My code is the bridge between high-dimensional theory and low-latency production.
"True innovation sits at the edge of theory and production. My goal is to compress the timeline between 'reading a paper' and 'shipping a feature,' delivering rigorous AI systems that survive user contact."
- Alan Kay