~/now ls

What I'm building now

Two projects at Filament Platform Inc., the company I founded in 2026.

~/now cat sensor.md

sensor · since Aug 2026

Privacy-first voice dictation for Linux. sensor runs as a background daemon on your own machine and transcribes speech without needing a GPU or anyone else's server.

[1] Runs everywhere Linux does: GNOME Wayland, KDE, X11 and a bare TTY, from one lightweight background daemon.
[2] 700ms on CPU-only hardware: End-to-end transcription latency with no GPU in the loop.
[3] One-command install, MIT licensed: Packaged for Ubuntu/Debian with clean uninstallation, and documented for end users.

~/now cat luminous.md

Luminous · since Jun 2026

Most useful information is trapped in pages that were never designed to be read by a machine. Luminous walks those sources and turns them into structured records: it decides what to look at next, extracts candidates, scores them, and reconciles the result against everything it already knows.

[1] Four-layer search pipeline: A central data contract holds the key information and propagates it across every layer, so each stage refines the same record instead of re-deriving it from scratch.
[2] Deterministic scoring, LLM judgment: The cheap, predictable work is done deterministically and the model is reserved for the parts that genuinely need judgment. Output stays consistent and the cost stays low.
[3] 101 named contacts in a single run: The current ceiling after moving to a stronger LLM backend — extracted, scored and reconciled without a human in the loop.

$ ./luminous --demo

A Luminous extraction run: the pipeline crawls a source, resolves candidate records across four search layers, and emits scored, named contacts into the shared data contract.

A recording of a live run is coming shortly. There is no public demo endpoint by design — the pipeline is metered, and an open endpoint would bill against real inference rather than prove anything a recording cannot.