Writing

Long-form pieces on quantum machine learning, scientific AI, the physics × ML interface, and the methodology questions that show up at all three intersections. Project notes from QMI Lab and AstroLLM land here as they ship.

Shorter informal pieces — paper reading notes, conference reflections, project-status updates — live separately at /writing/notes/ (feed).

  • Amounts, not relationships: what confidential transfers on the XRP Ledger hide

     · 23 min read  · XRP Ledger · Confidential Transfers · Zero-Knowledge Proofs

    XLS-96 lets a token holder keep balances encrypted on the public XRP Ledger. I built the core of a receivables facility on it and ran it on Devnet to see exactly where the line falls. It hides how much moves inside confidential form. It does not hide who pays whom, when, the a...

  • Unlearn the loop: what J shows that q hides

     · 21 min read  · q · kdb+ · J

    q has loops, and they work, so q alone will never make you stop writing them. J has loops too, but writing one means leaving its notation, and that feeling is the whole case for a short detour before q. Three ideas come back from it, and two of them break on the way home: one ...

  • The as-of join: what changes when the engine is built around one primitive

     · 21 min read  · q · kdb+ · Time Series

    pandas, Polars and DuckDB all have an as-of join; it is not q's secret. What changes in q is that aj comes apart into two primitives, group and bin, standing on a sort you are responsible for, and every edge case of the join is settled before you write a line. No timings, by c...

  • No headroom: what a null result on KX's q plugin actually measured

     · 18 min read  · q · kdb+ · Evaluation

    I set out to test whether KX's official Claude Code plugin for q measurably improves a frontier model's q output; on fifteen paired tasks I could not measure any improvement, and the more useful finding is why.

  • Array thinking, all the way to q

     · 11 min read  · q · kdb+ · J

    I am writing a short curriculum that takes an imperative programmer through the array-programming paradigm shift and lands them in q/kdb+, with a brief stop in J on the way, and every code example in it runs.

  • The post-quantum gap in Zig — shipping slh-dsa-zig

     · 20 min read  · Post-Quantum Cryptography · Zig · Open Source

    slh-dsa-zig is a pure-Zig implementation of SLH-DSA (FIPS 205), all twelve parameter sets, KAT-validated against NIST ACVP. What hash-based signatures are for, what Zig was missing, what this library does and does not yet claim, and where it goes next.

  • What five rounds of adversarial review taught me about my own design

     · 17 min read  · AI Agents · Trust · Architecture

    The four-axis design I argued for in Part 2 was not my first answer. It was my fourth. This is the record of how it got there — and the pattern in everything I got wrong on the way.

  • How Lodestar records what an agent believed

     · 21 min read  · AI Agents · Trust · Memory

    Four orthogonal lifecycle axes, the rule that nothing promotes itself, and the audit trail that falls out the other end.

  • The question my coding agent couldn't answer

     · 13 min read  · AI Agents · Trust · Memory

    Agents act, agents remember, but the standard agent stack rarely lets you reconstruct what the agent actually believed when it acted. This is a problem worth a name.

  • Three axes of cost — what a matched-resource baseline actually means in quantum machine learning

     · 9 min read  · QMI Lab · Quantum Machine Learning · Methodology

    Quantum machine learning has a comparison problem. Most published QML results compare quantum methods against classical baselines that have not been given a fair share of resources — and the field is starting to notice. This is the methodology I am holding QMI Lab's Pillar II ...

  • Quantum-safe custody — what changes for hardware-anchored and MPC signing when ECDSA falls

     · 12 min read  · Quantum Cryptography · Post-Quantum Cryptography · Digital Assets

    The planning baseline for Q-Day compressed in March 2026, and the question for institutional digital-asset custody is no longer when, but whether the cryptography can be migrated fast enough. Hardware-anchored signing and MPC have different migration paths — and one will be su...

  • Starting QMI Lab — three pillars, one question

     · 7 min read  · QMI Lab · Quantum Machine Learning · Machine Intelligence

    QMI Lab is an independent research lab studying intelligence, learning, and representation across classical and quantum computation — three pillars at three time horizons, one question underneath. Why I started it, how it is organized, and what is coming.

  • The model I wish I had as a graduate student

     · 6 min read  · AstroLLM · Scientific AI · Domain-Specialized LLMs

    AstroLLM is an open-source, retrieval-grounded language model for astronomy and astrophysics — designed around the workflows researchers actually use rather than the benchmarks frontier models optimize for. Why I started it, where it is, and what is next.

  • Three fields, one question

     · 4 min read  · Project Notes · Cross-Domain · Intro

    A notebook for working out loud across machine intelligence, quantum computing, and the digital-asset infrastructure those two fields are starting to reshape. Why I write here, and what is coming.