AI Architect
Investor × Web3 Researcher × WEB Marketer × Editor
I don't start with technology — I start with problems I deeply understand.
Using COA (Cognitive OS Architecture), a reasoning framework I've spent over a year refining through real-world practice, I structure problems before writing a single line of code. Claude Code then turns that structure into working products.
This isn't about domain expertise in any one field. It's a method: find a problem you have the judgment to evaluate, structure it with COA, build it with Claude Code. The products below span finance, real estate, and physical fabrication (3D printing and CNC) — different domains, same method.
"AI Architect — not someone who writes code for a living, but someone who uses AI to engineer solutions that matter."
Subscription SaaS for strategic analysis — powered by Cognitive OS Architecture, a proprietary reasoning framework built on top of Claude
Rather than prompting an LLM directly, AXIS-COA runs each session through Cognitive OS Architecture — a structured cognitive pipeline executing across four discrete phases: web research, situation structuring, assumption surfacing, strategy generation and strategy crystallization. Input your context, challenges, and goals; AXIS-COA handles the rest. Supabase handles auth and persistence; Stripe powers subscription billing with plan-based limits, backed by a FastAPI webhook layer for real-time plan sync. Hosted on Render.
Real estate investment analysis platform — combines GSI map tiles, COA reasoning, and real transaction data from Japan's MLIT to structure "what conditions would make this investment work"
Rather than a binary buy/don't-buy verdict, TERRAIN-COA presents the conditions under which an investment becomes viable — price negotiation targets, financing structures, exit strategy triggers. Real transaction data (Reinfolib), elevation/disaster risk (GSI), facility and station data (Yahoo! Local Search), and population trends (e-Stat) are combined with a Claude-powered COA analysis (Market → Asset → Strategy → Risk → Boundary).
Turn a single photo into a 3D model — for pets and children, with consumer-grade simplicity
Today, 3D printers and 3D modeling tools are powerful but built for professionals — most people who'd love a 3D-printed memento of their pet or child have no way to create the data. katachi closes that gap: upload one photo, get a free gray-mesh preview, and purchase a textured GLB/STL ready for printing. Pet/child photos were chosen as the entry point because the emotional motivation is strong and immediate — but the underlying goal is bridging the gap between "people who want something made" and "people who have a 3D printer."
Convert a photo into a CNC-ready relief — carved, not printed
🔗 github.com/ShibaInuChan/Relief-Forge
Where katachi builds form up with a 3D printer, Relief-Forge carves it away. Upload a photo and it becomes a grayscale depth map, then a relief transferred onto a flat plate or an existing STL surface — exported as B-spline IGES, CNC-ready G-code, and a preview PNG, ready to hand straight to a CNC router or mill. Same question as katachi — turning a photo into a physical object — solved through a different output.
Smaller tools built to solve my own day-to-day problems — same method, smaller scope.
US & Japan macroeconomic indicators with monthly AI-powered structural analysis
🔗 investment-dashboard-sooty.vercel.app
US & Japan macroeconomic indicators (CPI, yield curves, rate spreads) auto-fetched from FRED twice daily via GitHub Actions, with monthly Claude-generated structural analysis.
Daily automated price alerts for investment trusts, equities, crypto, and gold
Scrapes prices via BeautifulSoup4 and pushes a morning summary to LINE every weekday via GitHub Actions.
A MoneyForward-like personal asset tracker
🔗 asset-management-nu-seven.vercel.app
Traditional and alternative assets (crypto, gold) tracked separately, with monthly snapshots and chart visualization.
A Claude Code workspace for transforming research, meeting conversations, and structured thinking into audience-specific presentation decks.
Three presentation modes adapt the same knowledge for different audiences: Research (self-contained learning decks), Consulting (analysis and proposal decks), and Executive (decision briefs for leadership).
Two rendering methods are supported. Method A (html2pptx) builds slides from HTML/CSS rendered via a headless browser, enabling free-form layouts with five design presets. Method B (template-based) unpacks an existing PPTX into XML, replaces text and images, and repacks it while preserving the original design.
Supports multiple input sources—including standard prompts, Deep Research outputs, and PLAUD meeting recordings—which are processed through a three-phase workflow: Knowledge Design, Slide Design, and Implementation & QA.
Blockchain-based loyalty points for local communities — zero legal risk by design
Blockchain-based loyalty points for local communities, on Polygon — designed from the ground up to avoid securities law and token regulation risk, with a LINE bot interface for non-crypto-native participants.
Every product here started not with "what can I build?" but "what problem do I have — and do I actually know what a good answer looks like?"