Writing

Papers and essays on applied AI and the engineering around it: RAG systems, vector search, agentic tooling and MCP, real-time and procedural rendering, and composable commerce architecture. Several are as much about the limits of these tools as their uses.

Ebooks

The Exoplanet Catalogue

Three papers on rendering every system in the Open Exoplanet Catalogue as an explorable 3D scene: the system itself, the procedural pipeline that derives each planet's appearance from its recorded parameters, and what visualisation can honestly claim about worlds no one has seen. See the project.

  1. A Catalogue of Alien Worlds

    System descriptionJuly 202630 min read

    Every one of the Open Exoplanet Catalogue's 4,081 systems rendered as an explorable 3D scene, with each planet's appearance derived from its recorded parameters rather than painted by hand.

  2. Plausible Worlds from Sparse Parameters

    Method description and epistemic boundsJuly 202628 min read

    A classification pipeline maps sparse catalogue records to sixteen planet types, a procedural shader renders them, and a four-tier speculation gradient grades every visual decision from measured to fictive.

  3. From Data Point to Destination

    Survey and synthesisJuly 202620 min read

    Exoplanets are the most pictured objects in science that have never been seen. A survey of how that imagery gets made, from press art to procedural universes, and why disclosure matters more than realism.

The Roman Archive

Four papers on a retrieval-grounded reading interface to a multi-author corpus of Roman prose: the system itself, the limits of retrieval over a classical archive, reading Rome through century-old translations, and the ethics of giving the dead a voice. See the project.

  1. An Interface to the Roman Archive

    System descriptionJuly 202644 min read

    The full account of the system: a multi-author Roman corpus behind retrieval-grounded generation, with per-author scoping, bounded first-person personas, and a sculptural gallery that answers aloud.

  2. What RAG Can and Cannot Do for a Classical Archive

    Survey and synthesisJuly 202612 min read

    Retrieval-augmented reading across a multi-author Roman archive: grounding, per-author scoping, citation discipline, and the limits of texts that are edited, translated, and partial.

  3. Reading Rome in Translation

    EssayJuly 202611 min read

    The originals beside their public-domain English: Shuckburgh's Cicero, Casaubon's Marcus Aurelius, and the rest, with what each translation preserves and what it loses.

  4. Giving the Dead a Voice

    On synthetic personae for historical figuresJuly 202616 min read

    The ethics of animating Rome's figures as retrieval-grounded personas: bounded knowledge, scoped to each author's own words, and never inventing what they did not write.

An AI Knowledge Base

Two papers on a git-backed markdown documentation platform: the system itself, with workspace multi-tenancy and MCP access for language models, and the two-pass vector retrieval pattern that searches it.

  1. An AI Knowledge Base

    System descriptionJune 202619 min read

    A documentation platform that keeps markdown in git as the source of truth, adds a Notion-style editor over it, and exposes the corpus to language models over MCP. The design decisions that mattered, and the ones that turned into bugs.

  2. Adaptive Vector Search in Practice

    Technical noteJune 202615 min read

    A two-pass retrieval pattern built on Matryoshka embeddings: shortlist against a truncated vector, rerank against the full one. With measured costs showing it is slower today, and the reasoning for shipping it anyway.

The ǀxam Archive

Three papers on a retrieval-augmented reading interface to the Bleek–Lloyd ǀxam folklore corpus, the problem of restoring a sleeping language's sound, and what language technology can and cannot do for endangered-language archives generally. See the project.

  1. An Interface to the ǀxam Archive

    System descriptionJune 202649 min read

    A retrieval-augmented reading interface that makes the public-domain Bleek–Lloyd ǀxam folklore corpus readable for a general audience, grounded strictly in the notebooks.

  2. Restoring the Sound of a Sleeping Language

    Feasibility study and project proposalJune 202624 min read

    Why no tool can voice ǀxam clicks today, and a proposal for a community-led pronunciation layer built from the living sister language Nǀuu, still spoken by Ouma Katrina Esau.

  3. From Record to Resource

    Survey and synthesisJune 202647 min read

    What language technology can and cannot do for the documentary archives of endangered and sleeping languages, across cases from Kenya to Aotearoa to Cornwall, with ǀxam as the worked example.