I build research-shaped systems: online judges that understand teaching data, AI-agent infrastructure that can inspect and improve code, and memory-centric modeling experiments that try to make long-context reasoning cheaper, sharper, and more auditable.

My favorite work lives at the boundary where a mathematical claim has to become a running system: proofs, tests, architecture diagrams, telemetry, and a repository that another engineer can actually clone, compile, and reproduce.

💡 Engram Retention PyTorch research scaffold for budgeted long-context memory: RetNet recurrence, hashed Engram lookup, Block Attention Residuals, and milestone snapshots.

🚀 CodeNexus An AI-native online judge platform designed for school teaching, judging, integrity workflows, and educational data intelligence.

graph TB
    %% Styling Definitions
    classDef research fill:#0f172a,stroke:#38bdf8,stroke-width:2px,color:#f8fafc;
    classDef systems fill:#0f172a,stroke:#2563eb,stroke-width:2px,color:#f8fafc;
    classDef core fill:#090d16,stroke:#06b6d4,stroke-width:3px,color:#06b6d4,font-weight:bold;
    
    subgraph Research_Loop ["The Research & Formalization Loop"]
        A["Research Questions"]:::research --> B["Formal Assumptions"]:::research
        B --> C["Executable Prototypes"]:::core
        C --> D["Rigorous Tests & Ablations"]:::research
        D --> E["Docs & Proof Trail"]:::research
        E --> B
    end

    subgraph Applied_Systems ["Applied Systems Engineering"]
        C --> F["CodeNexus <br> (AI Education & OJ)"]:::systems
        C --> G["DNA-Memory / Evolver <br> (Agent Memory Scaffolds)"]:::systems
        C --> H["Engram-Retention <br> (Long-Context Models)"]:::systems
        C --> I["RuView <br> (WiFi Sensing & Edge)"]:::systems
    end

    %% Adjust layouts/spacing
    style Research_Loop fill:#070a13,stroke:#1e293b,stroke-width:1px,color:#94a3b8
    style Applied_Systems fill:#070a13,stroke:#1e293b,stroke-width:1px,color:#94a3b8

Layer	Technologies & Frameworks
🚀 Core Systems & Languages
🧠 Deep Learning & AI Research
💾 Infrastructure & Pipelines

"Simplicity is a prerequisite for reliability." — Edsger W. Dijkstra

• Executable First: Start with the actual running system, not a slogan or an empty abstraction.
• Ablated & Proven: Keep claims narrow and humble until comprehensive tests and rigorous ablations make them stronger.
• Falsifiable Architectures: Design architectures that can be actively inspected, systematically reproduced, and disproven.
• Evidence-Centric AI: Build AI features around solid empirical evidence, auditability, and governance workflows, not just chat wrappers.
• Documentation as Code: Treat system documentation and formal proof trails as core components, never as packaging after the fact.

host: XXY-CH@github
status: active
repositories: 19 public
focus_areas: [Engram Retention, CodeNexus, OpenMAIC, RuView]

[Languages by Repo Count]
JavaScript  ████████████████ 4
C++         ████████████ 3
Python      ████████████ 3
Rust        ████████████ 3
TypeScript  ████████████ 3
Go          ████ 1

• Email: cachoidxx@gmail.com
• GitHub Profile: github.com/XXY-CH