Full-stack + mobile engineer building product systems, developer tools, and AI-native delivery workflows.

Current focus:

• 🗞️ VeraMint - mobile news product focused on human-curated journalism, offline-first reading, and media literacy
• 📸 Picalyze - photography analysis platform for upload, analysis, and workflow exploration
• 🚇 qwer-q - typed, docker-first message queue with schema registry and dashboard
• 🔌 stead-core - vendor-neutral session standard for Codex and Claude Code interop

Current setup:

• Codex via GPT-5.4, mainly in the CLI, with the Codex app also in use
• Claude Code with Opus 4.6 and the frontend-design skill for UI work
• T3 in evaluation, but not yet a stable default lane

Live and runnable:

• 🗞️ VeraMint
• 📸 Picalyze
• 🚇 qwer-q
• 🖼️ passepartout
• 🌌 aurora
• 🧰 stead

Proof of motion:

• 2,005 GitHub contributions
• 183 PR contributions
• 251 tracked Codex threads
• 2.98B tracked Codex tokens
• 2,008 Claude history entries

• 🧰 stead - workspace orchestrator for terminals, ports, and context
• 🌌 aurora - European community platform with a Convex-first realtime core
• 🖼️ passepartout - photography-first export tool for framed image output and social-ready formats
• 🌐 edhor-me - personal site and portfolio with an AI-powered ask-me flow
• 🧠 Thinking Loop - deep research and idea-convergence workflow built around iterative agent loops
• 🤝 Fair Creator - transparency-first platform for creator-contractor relationships and fair work standards
• 🔏 Image Provenance - verifiable image authenticity that survives social media distribution
• 📡 Upstrio - European streaming platform exploring fair economics and distributed infrastructure

• Shipping current product work across VeraMint, Picalyze, and private systems that are part of my long-term product portfolio
• Building public tooling like qwer-q, stead-core, stead, aurora, and passepartout
• Running a Codex-first workflow for implementation and review, with Claude Code as the UI and frontend-design lane
• Keeping the public side runnable and inspectable while private systems stay visible through concrete product descriptions

My AI usage is structured, repo-bound, and phase-dependent rather than one generic chat stream.

• Planning and task shaping stay inside the build loop. In Codex thread openers alone, planning-related prompts show up 65 times, documentation 58, bugfix 49, review 35, research 28, and testing 23.
• Frontend and product-shaping work are a major slice of the loop. Frontend/UI signals show up 98 times in Codex thread openers and 489 times in Claude transcript user messages.
• Parallel execution is part of the normal operating model. Worktree-specific Codex prompts appear 19 times, explicit agent prompts 9 times, and non-main agent roles account for 32 threads and about 699M tracked tokens.
• The setup is opinionated rather than scattered. Codex is the main delivery lane, and Claude Code is the UI and frontend-design sidecar when visual work matters.
• The work stays project-centered. Codex-heavy work clusters around VeraMint, stead-core, passepartout, Aegis, and Nexum. Claude-heavy work clusters around Picalyze, VeraMint, qwer-digest, qwer-q, and Upstrio.

• Claude carries the high-volume exploration layer. January 2026 alone has 1,512 Claude entries, with dominant signals in bugfixing, planning, testing, and brainstorming.
• Codex carries the heavier repo-bound execution layer. The current setup is GPT-5.4, mainly through the codex CLI, with the Codex app also in use.
• Claude Code is the frontend and UI lane. The current setup there is Opus 4.6 together with the frontend-design skill.
• T3 is in evaluation, but not yet part of the stable default workflow.
• March 2026 stays Codex-heavy: 76 threads and 1.22B tracked tokens, while Claude drops to 14 entries and mostly reads as a frontend/UI sidecar. The top Codex signals in March are frontend, docs, brainstorm, and plan, which reads like active product and tooling buildout rather than passive chat.

• 2026-W06: Claude-heavy phase with 221 entries across 29 sessions and almost no Codex activity.
• 2026-W08: Codex takes over with 110 threads and 738.7M tracked tokens while Claude is down to 21 entries.
• 2026-W10: Codex remains dominant at 53 threads and 717.0M tracked tokens; Claude is only 12 entries.
• The pattern is consistent: early 2026 starts as high-volume Claude exploration, then shifts into Codex-centered delivery, parallel execution, and longer implementation sessions.

• AI: Codex, Claude
• Current setup: Codex via GPT-5.4, mainly in the CLI, plus the Codex app
• UI setup: Claude Code with Opus 4.6 and the frontend-design skill
• In evaluation: T3
• Workflow: git, git worktree, gh, PR review loops, multi-agent execution
• Product stack: TypeScript, Bun, React, Next.js, Astro, Expo, Convex, PostgreSQL
• Delivery and QA: Playwright, Biome, CI workflows, Vercel, Netlify
• Systems work: Go, Rust, Docker

Snapshot as of 2026-03-22.

• 2026-W06: Claude 221 entries across 29 sessions, Codex 1 thread
• 2026-W08: Codex 110 threads and 738.7M tracked tokens, Claude 21 entries
• 2026-W10: Codex 53 threads and 717.0M tracked tokens, Claude 12 entries
• 2026-W12: Codex 7 threads and 16.5M tracked tokens, Claude 1 entry

Codex

Claude