The Plutarc Treatment
A few things have shipped across the Valeon stack worth writing down.
GigFin and OEMI both got the Plutarc treatment. Both codebases were thrown out and rebuilt from scratch — same UI language as the rest of Valeon, same design philosophy, nothing inherited from the old implementations that didn't deserve to survive. GigFin now ships with a full Convex backend baked directly into the image using self-hosted Convex, deployed as a dependent service through Docker Compose. That matters because it unlocks scheduled and cron jobs natively, which meant recurring payments could be implemented properly rather than bolted on. They're in. Tax reports are in too — generated on the fly, jurisdiction-compliant, not tax advice but accurate enough to serve as a working guide for anyone who needs to know where they stand. Every aspect of the product is now documented in the repo, written for users, not developers. OEMI's redesign brought the interface into line with the broader Valeon stack, and identity submissions have been meaningfully improved — there's now a full TipTap editor with inline previews for LaTeX equation submissions, which changes the submission experience considerably when the work involves notation. A full prerendered path has been implemented, similar to how Valeon handles it, so pages load fast without requiring a complete rebuild or redeploy of the project. Coming shortly, OEMI will gain AI-assisted validation of submissions alongside assisted prefill of certain fields — description, citations, sources — reducing the friction of a submission without compromising the rigour of what gets through.
Wavegram has been quietly improving. The forward and backward processes have both been overhauled, with a 2048-point FFT window now available alongside the existing options. The reconstruction side is the more interesting change — phase is now recovered statistically rather than assumed, which means the reconstructed audio is lossy in the honest sense of the word, but substantially more faithful than what came before. A built-in audio recorder has been added directly to the page, and playback now runs through Valeon's wavesurfer.js powered audio player, so sampling and previewing a reconstruction is immediate without leaving the tool.
VocaSync's synthesis engine has doubled its concurrency, which is the kind of change that only becomes visible under load but matters considerably when it does. Phonetic pronunciation inputs are now supported across both the dashboard and the API — users can supply pronunciations directly, and VocaSync handles the reconciliation against the speech text in the backend before anything reaches the synthesis step. The original transcript is left untouched throughout, which keeps it clean for alignment and translation downstream. The WordPress plugin has landed. It is fully built, tested, and sitting in the public repo. Submission to the WordPress plugin catalog is in progress; manual installation works in the meantime.
Valeon itself has not been left out. The platform received its own Plutarc treatment in the form of comprehensive documentation covering the author dashboard and every surface and piece of functionality it exposes — future authors now have a single place to go to understand how to use the platform rather than piecing it together. This came in tandem with date and year delimiters in the synthesis pipeline, which gives it the context to distinguish between a number, a date, and a reference to a year when converting post content to audio. The pipeline has become genuinely mature at this point — it can handle more or less any style of post thrown at it. Tables remain the exception. I haven't figured out how to handle those yet.
AI-supported features have been extended to tags. Posts now receive automatic tag handling, with AI-suggested tags generated from post content — a small addition that compounds over time as the content archive grows.