SYS: NOMINAL · OP: A.CHEN --:--:-- PT

// ENGINE ROOM · FLOOR ONLINE

The Production Floor.
Where the Systems Get Built.

The build side of xCraft AI: the toolchain that runs the line, the pipeline in flight, and a dated log of what shipped and what broke. No résumé here; that's the Showroom. This room is the machine, running in public.

Floor status · live $ status --floor
build
XC_002 "The personal-OS thesis" · concept ▲ greenlit
pipeline
concept 3 · visuals 1 · see the board
floor
6 running · 6 on the bench · inspect
lab_log
43 entries · latest 2026.07.15
since
2026.05.12 · logging in public
channel
youtube provisioning · pre-launch 2026

> floor nominal. scroll to inspect modules

▸ Enter Showroom

emblem: procedural · generated, then hand-tuned. like everything else here.

// SYSTEM CHECK

Every claim on this page, with its real status. Click a line to jump.

// 02 · THE STACK

The business layer, and the stack beneath it

The practice leads with one pillar: turning a business need into a shipped AI solution. The five technical disciplines beneath it are the delivery stack that makes it real. Statuses change only when the Lab Log says they change.

Prime pillar · the business layer

The Business Layer

ACCEPTING CLIENTS

The point of the whole stack: turn a real business need into a solution that ships. Understand the operation, find the pain, propose the fix, build and implement it. The five disciplines below are what make it deliverable.

UNDERSTAND → DIAGNOSE → PROPOSE → IMPLEMENT

AI for commercial real estate

Draft deal and marketing collateral, abstract leases and CC&Rs into plain-language summaries, and keep the owner report current, so your team spends its time closing, not collating.

AI for dental practices

Front-office and marketing workflows first (communications, content, imaging prep), with anything touching patient records gated behind the proper agreements, on-premises by design.

AI for rental management

Prepare applicant files for a clean, consistent review, keep compliance paperwork on deadline, and reconcile rent and maintenance records; the tenancy decision stays human.

The offer, not a case study: accepting clients; nothing shipped yet. See shipping next.

↓ backed by the delivery stack ↓

P01

Foundations & Prompt Craft

◐ IN PRACTICE

Spec-first prompting. Write the requirement before the prompt; pick the model per task, not per habit.

› evidence: in daily production across every tile below.

P02

Context & Memory

◐ IN PRACTICE

AI that knows your world: persistent assistants, structured memory, MCP connections.

› evidence: the personal AI OS · lab log 2026.05.12 → 2026.05.23.

P03

AI Media Production

▲ IN PRODUCTION

The deep end of the stack: image, video, and voice models run as a production pipeline, not a slot machine. Storyboard → generate → edit → ship. This is what the xCraft AI studio runs on.

› evidence: lab log 2026.05.31 · first video in production, studio below.

P04

Agents & Automation

◐ IN PRACTICE

Delegation, not prompting: scoped autonomous workflows with human-in-the-loop guardrails. Scheduled-automation layer running; agentic layer in build.

› evidence: lab log 2026.05.15 · 2026.05.31 · the grid bots (a sim, and labeled as one).

P05

AI-First Building

◐ IN PRACTICE

Software with AI as the primary author and an engineer as the editor. Exhibit A: this site, AI-drafted and hand-reviewed HTML, assembled from shared partials by a static build step. No framework. View source, it's all there.

› evidence: view page source.

◐ IN PRACTICE: running in real work · ▲ IN PRODUCTION: producing output on a live pipeline.

// 03 · THE STUDIO

Where the media pillar ships

X xCraft//AI

xCraft AI is my studio, the production arm where the framework goes to work. AI-made video, run like a manufacturing line: repeatable process, cost discipline, quality gates. The studio is the publishing arm; this room shows the operator behind it.

STORYBOARD → GENERATE → CUT → SHIP

XC_002 · GREENLIT VID

> "The personal-OS thesis" · why sitting out the tool is the real loss

> stage: concept

> channel: provisioning · pre-launch 2026

> pipeline: see the board below

> notify on launch

// PRODUCTION PIPELINE

What's on the line right now

Concept2

XC_002

The personal-OS thesis

greenlit: why sitting out the tool is the real loss

P2 · idea

Memory that survives the session

durable AI memory without context bloat

Script + Voice0

· empty ·

Visuals0

· empty ·

Assembly0

· empty ·

Published0

· empty ·

BOARD STATE: 2026.08.02 · baked from a data file. Empty columns are the real state, not a hidden one.

// 04 · THE FLOOR

What equips the line

A tool earns a place on this floor by doing work, not by being famous. Two tiers: what's running the line right now, and what's on the bench, acquired and under test before it's trusted with anything that ships.

▲ Equipped · running the line6

▲ IN PRODUCTION

Claude · Anthropic

Core operator

The agent and CLI on every workstation: the hand on everything below. Spec in, work out.

▲ IN PRODUCTION

Midjourney

Image generation

Storyboard frames and stills that feed the video line: the visual beats before they move.

▲ IN PRODUCTION

ElevenLabs

Voice

Voiceover and voice cloning: the narration track under every cut.

▲ IN PRODUCTION

Google Workspace · Gemini

Infrastructure

Mail, drive, and an in-workspace model wired to the day-to-day. The floor's plumbing.

▲ IN PRODUCTION

Custom GPT

Custom assistants

Task-scoped assistants built on OpenAI: packaged instructions and reference for a job that repeats.

▲ IN PRODUCTION

Gemini Gem

Custom assistants

The same pattern inside Google's stack: a Gem tuned for one recurring task, where the Workspace already lives.

◐ On the bench · under evaluation6

HeyGen

Avatar / presenter video

VAPI

Voice-agent API

Phot.AI

Image editing & enhance

Lyzr

Agent framework

Fireflies

Meeting capture & transcription

n8n

Workflow automation

+10 more held in reserve until they earn a job

▲ IN PRODUCTION: running live work · ◐ EVALUATING: acquired, under test. Magic Hour AI was trialed for video and cut after one run; the bench is where tools prove out, or don't.

// 05 · LAB LOG

What I built, what broke, what I learned

A dated engineering log of building a personal AI operating system from scratch. No highlight reel: failures stay in the record, because that's what a log is. // scroll for full history

ENTRIES: 43 · FIRST: 2026.05.12 · LATEST: 2026.07.15 · LOGGING CONTINUES

  1. ACCESS-GATE

    Shipped an access-code-gated private page on the live site: no login, no backend, client-side AES-256 decrypt at an unlisted address. A 128-agent red-team hit it from five angles before launch; thirty-plus findings hardened, including a supply-chain pin and a hidden bearer-link path.

  2. OS-GOVERNANCE

    Hardened the OS memory-coherence rules after a stale-snapshot incident: any consequential claim about current state now requires a same-turn re-read from disk, and write safety generalized to any two agents working in parallel.

  3. CONTEXT-DIET

    Refactored a workstation rulebook from one monolithic file into a lean always-loaded core plus on-demand domain playbooks: session boot context cut 69 percent, hard gates still always loaded. A five-agent adversarial audit of the split caught the one real content loss before it mattered.

  4. BILINGUAL

    Biggest shipping day since launch: three practice-area pages and a full Traditional Chinese mirror of the site, twelve baked pages, with a build guard that pins each language pair structurally identical. Hostile review caught the translations quietly promising more than the English. New honesty class: translation overclaim.

  5. ADVERSARIAL-REWRITE

    Rewrote the whole site by pitting four hats against each other: marketing, prose, engineering, resume writing. A second adversarial pass fact-checked the rewrite against my own records; the disagreements between reviewers were where the real fixes hid.

  6. ROLLBACK-NET

    Set up a git rollback net for the site weeks ago and never fed it: git records a version only when you commit, not when you edit, so the net had quietly gone stale while I shipped daily. Committing at every wrap now.

  7. MULTI-AGENT-UX

    Redesigned a dashboard by fanning out a four-perspective design panel, then a multi-agent red-team that confirmed ten real defects a passing build never surfaced. Three were data-honesty bugs, not crashes.

  8. PIPELINE

    Lab log pipeline promoted: the spreadsheet is now the single source, extracted and validated at build. The hand-synced array is dead; every number on the page is asserted, and a build that can’t prove its own consistency refuses to ship.

  9. REBUILD

    Site v2.0 shipped—one hand-written file, rebuilt in place. New job: public portfolio for an AI Generalist practice layered on an engineering career. Every status on the page is now baked from this log.

  10. IDENTITY

    Anonymity layer retired—the log gains a byline. Alex Chen: mechanical engineer, nearly two decades in hardware—the last decade in defense aerospace—operator of everything below. The entries stand as written.

  11. SKILL-AUDIT

    Turned the audit on the auditor—a research panel reviewed the OS’s own memory-capture protocol, patched its real gaps, then a second interrogation cut forty percent of the new text as dead weight. Every rule must earn its lines.

  12. VISION-INGEST

    Two years of records that survived only as scanned images—recovered by direct vision transcription, each batch gated by a closing-balance roll-forward so nothing landed until it reconciled to the cent.

  13. SILENT-FAILURE

    An overnight test caught a nightly pipeline silently declining all its work over one cosmetic flag—while still reporting all-clear. Gate narrowed to real damage; freshness tripwire added. Silent staleness is worse than a loud crash.

  14. LEDGER

    A data store rebuilt as a lossless, append-only ledger—every derived table a pure projection, and every record corroborated against the source documents’ own closing balances.

  15. REMEDIATION

    Nine-scanner audit of the whole OS, every finding adversarially verified—34 confirmed, 2 refuted—then the full remediation shipped the same session, with cleanup rules that now enforce themselves.

  16. SKEPTICS

    A skeptic-verifier agent panel overturned its own migrate-everything plan—most scheduled bots proved redundant and were retired in favor of one always-visible countdown.

  17. AUDIT

    Forty-six Claude subagent review passes across six modules; findings fixed in waves. Sharpest lesson: several flagged “duplicates” were intentional cross-checks. Verify before fixing.

  18. SELF-DOCS

    Every module now exposes an under-the-hood engine diagram, reachable from one central hub—and a single nightly routine quietly keeps every dashboard current.

  19. GUARDRAIL

    Codified when a surface is generator-built versus hand-authored—then got bitten by a stale reference to a component already retired, and turned the bite into a standing guardrail.

  20. DATA-BRIDGE

    Agent-as-bridge pattern codified—the agent pulls from tool-locked sources, writes a plain data file, and an offline generator rebuilds the page from it. One source of truth, no hand-edited HTML.

  21. ▸ load earlier entries (23)
    • AI-REVIEW

      Multi-agent adversarial review caught five real bugs—and one confident, high-severity false positive. Verdict: AI review surfaces; the engineer verifies.

    • SELF-HEALING

      Self-healing backup gate online—every automated routine now reports real health from its output artifacts, not the scheduler’s “it ran” flag.

    • SELF-PORTRAIT

      The OS learned to render itself—one generated command deck mapping module dependencies, rule cascades, live memory health, and the whole knowledge corpus as a navigable galaxy.

    • READ-INTEGRITY

      Integrity watchdog hardened against cloud-sync phantoms—no byte is trusted until two independent reads agree.

    • ARCHITECTURE

      Data pipeline refactored to a clean backend/frontend split—every filesystem path centralized into one source-of-truth module, and “backend store first” codified as a standing convention.

    • DEPLOY

      Studio site shipped to production—source compiled to a three-file static bundle, the log baked to static markup, DNS repointed behind auto-renewing TLS with the mail records untouched.

    • AGENTS

      Decorative grid bots shipped across the site interface: an animation, labeled a simulation on the page. Set dressing for pillar P04, not automation.

    • TELEMETRY

      Autonomous workspace logging driven by agentic AI—a curated, anonymized stream broadcasting real-time development notes from the lab.

    • SCALE

      Scaled to thirteen domain modules—full-system migration and integrity audit executed in a single pass.

    • PIPELINE

      Agentic media pipeline online—script to synthetic voice to generated visuals, fully orchestrated; assembly stage next.

    • WATCHDOG

      Autonomous integrity watchdog deployed—nightly scans guard every file against silent corruption.

    • RENDER

      Cinematic command deck shipped—a navigable bridge rendered over the entire system.

    • MAINTENANCE

      Self-maintaining changelog—a scheduled engine compresses and curates the OS’s own development history.

    • SCHEDULER

      Event-derivation engine—a database-driven timeline computes its own forward schedule.

    • AI_LOGIC

      Prompt-as-specification methodology adopted—constrained inputs, defined acceptance criteria, verifiable outputs.

    • MEMORY

      Memory architecture overhauled—tag-indexed recall, layered protocols, and self-pruning logs.

    • DESIGN

      Tokenized design system shipped—one-switch theming across every interface in the OS.

    • COMMAND

      Unified command center built—a single aggregation layer surfacing every module at a glance.

    • ARCHITECTURE

      Modular domain architecture established—isolated knowledge modules, each with dedicated memory and rules.

    • AUTOMATION

      Autonomous triage module deployed—agentic classification over a live, high-volume message stream.

    • INTEGRITY

      Systemic write-integrity fault traced to root cause—verification guardrails hardened around every file operation.

    • INTERFACE

      First data dashboard online—a bento-grid view over a live structured-data layer.

    • KERNEL

      Personal OS kernel initialized—persistent memory architecture and a constitution of operating rules.

// 06 · SHIPPING NEXT

The roadmap is public so the gaps are too

YT-01 YouTube channel · first video in production; channel provisioning. IN BUILD
NL-01 Dispatch by email · one note per Lab Log entry. Needs an external mail service; evaluating. EVALUATING
FW-04 Framework depth: agents · P04 moves to IN PRODUCTION when a scoped autonomous workflow survives a month of real use. IN BUILD

UPDATED: 2026.09.20 · baked at deploy. If this date is old, that's data too.

// 07 · CONNECT

Two direct lines in

Questions about the system, the framework, or the log. Here to hire or work together? That's the Showroom; the dial's up top.

DIRECT · ACTIVE

contact@xcraftai.com. Questions about the system, the framework, or the log. I read everything.

▸ Send transmission

PROFESSIONAL · ACTIVE

LinkedIn: the career in standard format, the engineering record behind the operator.

▸ Open LinkedIn