Systems engineer and memory architecture researcher. I design deterministic, low-level associative memory engines in Rust that turn raw byte streams into cryptographic Merkle DAGs, and I have a strong preference for mathematical invariants over opaque neural black boxes.

Most modern cognitive systems rely on multi-gigabyte vector databases, floating-point embeddings that drift with every checkpoint, and fragile prompt scaffolding. I build zero-vocabulary state machines where memory is byte-exact, compression is calculated online, and active working memory is bounded by provable temporal reachability.

Day to day:

• Rust (2024 Edition) — memory arenas, zero-copy byte streams, lock-free concurrency, deterministic data structures
• Merkle DAGs & Cryptographic Interning — SHA-256, BLAKE2, bottom-up concept emergence from raw factory bytes
• ACID Embedded Storage — single-transaction durability via redb, memory mapping via memmap2
• Information Theory & Surprise — online surprise calculation ($S_{\text{static}}, S_{\text{growth}}$), streaming greedy folding $O(N)$
• Graph & Co-occurrence — sparse PPMI matrices, associative recall, biased random walks

Opinions I hold quietly:

• If a system cannot explain its internal state without querying another LLM, it is not a memory engine—it is a cache with opinions
• Vector databases are often an expensive workaround for the lack of a proper emergent grammar
• Generational GC without sweep ($K=3$ temporal root reachability) is all the forgetting an agent needs
• 100 passing deterministic tests beats a 1,000-line prompt every single day
• If state is not byte-exact and reproducible, debugging it is just guesswork

axmg — Universal Merkle DAG Associative Memory Engine for Rust. Embedded, deterministic, zero-vocabulary associative memory engine. Replaces heavyweight vector databases with an online cryptographic DAG, dynamic working memory ($K=3$), online surprise metric, and sub-graph PPMI recall. Published on crates.io/crates/axmg and GitHub Packages.

axiom — Foundations of associative cognitive structures, byte-level tokenization, and emergent linguistic topologies.

axmv2 — Explorations in multi-agent memory synchronization and temporal graph evolution.

Currently formalizing Model Context Protocol (MCP) bridges so autonomous agents can plug directly into deterministic associative memory without middleware overhead, and exploring zero-alloc arena compaction for long-running streaming daemons.

Open an issue or start a discussion on any of the repositories above. For architectural inquiries or collaboration on deterministic memory engines, reach out via GitHub or email.

Deterministic · Byte-exact · Built with Rust