I build automated systems that run themselves — and businesses that run on them.
I’ve spent 27 years shipping production software — the last several building AI-native systems that run six real businesses without me in the loop. I design, build, and operate automated pipelines for content, real estate, e-commerce, and publishing — mostly solo, in Python and Go, live on a Linux VPS, unattended by design.
I don’t build demos. Everything below is in production, generating real outcomes, right now.
17 years continuous operation · 110% sales lift in 6 months · 150–300 qualified leads/week · Tools other builders adopted on their own
Current AI Stack
CLI Orchestration: Claude Code · Codex · Qwen · Gemini · OpenCode · Pi · gws CLI · CLI Anything
AI APIs: OpenAI · DeepSeek · Anthropic · Google · Groq · OpenRouter · Fal
Agent Governance: Four tools I run together on every agent session — not because any one is a silver bullet, but because the combination compounds: gate, verify, compress, repeat. Governator gates every tool call before it runs. Assayer verifies AI-generated output before it ships. RTK compresses command output 60–90% before it ever reaches the model. Context Mode keeps raw tool output out of the conversation entirely, processing it in a sandbox and surfacing only the derived answer. Run together, this is the difference between a multi-agent pipeline that’s affordable to leave running and one that quietly burns through budget. Also worth adding: OmniRoute — a free AI gateway routing one endpoint across 290+ providers and 500+ models with quota-aware auto-fallback, so a rate limit on one provider never stalls the pipeline.
Featured Projects
Governator
A Go-based supervisor sitting above coding-agent CLIs — Claude Code, Codex, GLM, OpenCode, Pi — turning file scope, command authority, and budgets into deterministic checks around every agent run. No tool call executes ungoverned: a separate validator and merge gate decide whether work ships, not the model itself, and OS-level containment (Landlock, network namespaces, cgroups) backs the guarantees at the kernel level instead of trusting the backend’s word.
It’s also a real cost-control layer, not just a safety one. RTK-aware prompting and command normalization compress command output before it reaches the model, a YAGNI-biased prompt injection keeps diffs small, circuit breakers route away from a failing provider instead of burning tokens on doomed retries, quota-aware routing avoids sending jobs to a backend about to hit a cap, and a hard spend cap halts a run before it becomes a surprise bill. GitHub
Assayer
A verification and trace layer for AI-generated output: verify → quarantine → trace. Built because “trust the model’s output” isn’t a QA strategy — every result is checked against configurable rules (schema, length, no boilerplate, no placeholder text, language sanity, dedup) before it’s allowed downstream, and anything that fails gets quarantined for review instead of shipped.
It runs equally well on an interactive coding agent’s output or a fully unattended API pipeline — it’s a plain library call, not something that requires an agent to be present. A durable local outbox means quarantine evidence survives even if the database write fails, and its pass/fail history feeds back into Governator’s routing decisions, so a backend’s real output quality — not just its uptime — shapes which one gets picked next. GitHub
Cookie Pipeline
One photo in, five platforms out, zero humans in between: photo drop → AI script → Thai TTS voiceover → vertical video → WordPress + TikTok + Instagram + Facebook + YouTube, fully automated.
Tarotsmith Network
17 years of continuous, revenue-generating operation (2009–present) — one of the longer-running AI-assisted content operations of its kind. AI-powered article generation, weekly horoscopes, a multi-agent writing pipeline. Live at tarotsmith.net.
Sawasdee Realty
An automated lead-gen system producing 150–300 qualified Bangkok leads a week — Python scraping, AI scoring, and multichannel distribution across TikTok, portals, and Line OA, the kind of work most agencies still do by hand.
Zone31
A multi-tenant AI divination platform — tarot, oracle, and dream interpretation — with a built-in safety/crisis layer and cost governance across every LLM surface. Live at oracle.zone31.com.
MonPaga / Waking Cup
Review aggregation and e-commerce automation that drove a 110% Lazada sales increase in 6 months for a Thai marketplace seller with zero in-house tech team.
Personal AI Tools
Solo experiments at the edge of what AI pipelines can do. Built for myself — published because other builders asked.
second-brain-interviewer
A Telegram bot that interviews you by voice. Dr. Soma — a psychoanalytic AI character — asks structured questions across 9 categories (values, fears, mental models, working style, history) and stores your answers directly into a personal second brain. State persists across sessions. ~$0.01/exchange, ~$0.50 for the full interview. GitHub
compass
Four questions, once a day. AI synthesis delivers one heading — a single direction for the session. Self-calibration ritual, not productivity theater. GitHub
future-self-letter
Claude Opus writes a letter from your future self, sealed for 90 days, delivered via Telegram on schedule. Forces the question: did you do what you said you would? GitHub
second-brain-brief
Daily morning brief — queries a Supabase vector store, synthesizes with Gemini Flash, arrives by email before you open your inbox. The second brain reports to you, not the other way around. GitHub
Personal Podcast Network
Eleven shows running on the same engine, covering every day of the week (some biweekly) — Gemini Flash script generation, multi-speaker TTS audio, Cloudflare R2 hosting, private RSS. No browser automation. No external services. Runs unattended.
Tubecap (Monday) — Digests your favorite YouTube channels and podcasts into a weekly audio recap. It’s a podcast about what happened recently on your favorite shows. GitHub
The Invisible Fire (Tuesday, odd weeks) — Hermeticism, Gnosticism, and the esoteric traditions beneath Western civilisation. Two hosts, eight rotating guest perspectives, 14-day source window across the best podcasts, blogs, and channels in the field. GitHub
The Shadow Cabinet (Tuesday, even weeks) — Psychology, narcissism, parapsychology, and the contested frontiers of the human mind. Clinical, depth, evolutionary, and frankly fringe. Eight rotating guest lenses including a Dark Triad strategist. GitHub
Mindstorm (Wednesday) — Mines the second brain: recent memories, flagged contradictions, unresolved decisions. Rotating hosts remix it into a full episode. Not just a recap but a remix. GitHub
The Labyrinth (Thursday, odd weeks) — Narcissistic abuse and CPTSD survivor education, with expert voices interrupted and reframed by two recurring personas who perform narcissistic traits rather than just describe them.
The Cartomancer (Thursday, even weeks) — Biweekly digest across the Tarotica publishing network — card by card, site by site, tradition by tradition. Two hosts plus a rotating guest lens.
The Stack (Friday) — GitHub Trending, HuggingFace papers, 14 curated RSS feeds. Weekly AI/builder digest through the lens of: what can we actually build with this?
The Crucible (Saturday, even weeks) — Alchemy and life-extension science, from historical occult practice to NAD+ trials. Self-improvement and health optimization through an esoteric lens.
High Strangeness (Saturday, odd weeks) — UAP disclosure taken seriously from every angle: FOIA documents, declassified encounters, and the non-human intelligence hypothesis. Four hosts, four different epistemic commitments.
The Hidden Curriculum (Sunday) — A 14-principle curriculum on the worldview, literacies, and institutions elite schools quietly train into future owners and operators.
The Distilled Mind (Sunday) — A 19-episode curriculum distilled from 400 book reviews. Two hosts: The Researcher and The Pragmatist.
personal-podcast-engine — The shared base class. Subclass it, implement fetch_sources(), get a full episode. Works with any input. GitHub