What I'm doing now
The honest snapshot. What actually has my attention this month, not a backlog of everything I've ever touched. Inspired by Derek Sivers' /now page idea.
The one thing
Mimir's native terminal emulator is the active build. Phase 8 is active piece by piece: rich-media decode/render, Kitty graphics lifecycle (upload, place, z-index, animation playback), basic sixel pixel decoding, shell-integration command regions (OSC 133), and a standalone mimir-term host seam with config-file foundation. Seven phases shipped with golden fixtures and VTE-as-oracle differential tests. ~55–60% through the full native migration; VTE remains default until Phase 9 proves native safe. Baseline audit done Jun 22; P0 next: throughput profiling + IME. App icons shipped. Personal project; may go public after the v1 migration is complete.
Also live on the bench
- Neuro-symbolic worker classification. Canada/Ontario V0 shipped. CLI questionnaire runner, factor-by-factor audit trail, JSON output. Joint inference combining legal rule factors with neural embeddings. #1 next-build candidate, $0 MVP, 4 weeks to demo. Research feed active with five new sibling papers (tax-law neuro-symbolic, LePREC, Rule Violation Score, CyberCane, SkillGraph).
- Nüwa-RECODED (private personal build). Clean-room reimplementation of my own agent CLI, built from zero, ~414k LoC plus ~543k LoC of tests, ~25.5k tests green across 1,450+ test files. Replaces the old Hermes fork. The fork vs redo are in the past; the clean-room rewrite is what matters now.
- Cold pitch campaign. 14 pitches sent June 19 to HN companies for $25–50K contracts. Waiting on replies.
- Blackreach v5. Autonomous browser agent, maintained as the daily driver for real research collection.
The frame
The throughline across everything I build: ship the tool, then publish what breaks. I build agent infrastructure that survives the real run, and the research that finds where the edges are. Local-first by default, my own hardware in Toronto, cloud only when a training run earns it. Tools I own end to end.
I'm self-taught and building toward a remote AI/ML role that makes a US move possible, but the move right now is to ship more first: the native terminal, the neuro-symbolic demo, and the private agent-CLI rewrite at parity. That's the unlock, not a detour from it.
Latest auto-update — 2026-06-24
Auto-draft. The phnix-dev-now-page cron should be wired to consume approved
versions of this draft instead of regenerating from scratch each time.
Building
(no active projects with status update)
Shipping
(no GitHub commits in window)
Reading
0 new findings from the Rynvyr research scanner.
Targeting
(manual — only you can fill this in)
Latest auto-update — 2026-06-25
Auto-draft. The phnix-dev-now-page cron should be wired to consume approved
versions of this draft instead of regenerating from scratch each time.
Building
(no active projects with status update)
Shipping
(no GitHub commits in window)
Reading
5 new findings from the Rynvyr research scanner.
Targeting
(manual — only you can fill this in)
Latest auto-update — 2026-06-27
Auto-draft. The phnix-dev-now-page cron should be wired to consume approved
versions of this draft instead of regenerating from scratch each time.
Building
(no active projects with status update)
Shipping
(no GitHub commits in window)
Reading
5 new findings from the Rynvyr research scanner.
Targeting
(manual — only you can fill this in)
Latest auto-update — 2026-06-27
Auto-draft. The phnix-dev-now-page cron should be wired to consume approved
versions of this draft instead of regenerating from scratch each time.
Building
(no active projects with status update)
Shipping
(no GitHub commits in window)
Reading
5 new findings from the Rynvyr research scanner.
Targeting
(manual — only you can fill this in)
Latest auto-update — 2026-06-28
Auto-draft. The phnix-dev-now-page cron should be wired to consume approved
versions of this draft instead of regenerating from scratch each time.
Building
(no active projects with status update)
Shipping
(no GitHub commits in window)
Reading
5 new findings from the Rynvyr research scanner.
Targeting
(manual — only you can fill this in)
Latest auto-update — 2026-06-29
Auto-draft. The phnix-dev-now-page cron should be wired to consume approved
versions of this draft instead of regenerating from scratch each time.
Building
(no active projects with status update)
Shipping
(no GitHub commits in window)
Reading
5 new findings from the Rynvyr research scanner.
Targeting
(manual — only you can fill this in)