██████╗ ██████╗ ███╗ ██╗███████╗ ██████╗ ██╗ ██╗ ██████╗ ██╗ ███████╗███████╗
██╔══██╗██╔═══██╗████╗ ██║╚══███╔╝██╔═══██╗██║ ██╔╝██╔═══██╗██║ ██╔════╝██╔════╝
██████╔╝██║ ██║██╔██╗ ██║ ███╔╝ ██║ ██║█████╔╝ ██║ ██║██║ █████╗ ███████╗
██╔══██╗██║ ██║██║╚██╗██║ ███╔╝ ██║ ██║██╔═██╗ ██║ ██║██║ ██╔══╝ ╚════██║
██████╔╝╚██████╔╝██║ ╚████║███████╗╚██████╔╝██║ ██╗╚██████╔╝███████╗███████╗███████║
╚═════╝ ╚═════╝ ╚═╝ ╚═══╝╚══════╝ ╚═════╝ ╚═╝ ╚═╝ ╚═════╝ ╚══════╝╚══════╝╚══════╝
Solo founder @ Yutro — AI automation for e-commerce AI systems architect · generative media pipelines · agents and RAG · GEO/AEO engineering
I build AI systems that have to run in production, carry traffic, and survive being maintained by one person. Everything below is deployed, not prototyped.
AI AUTOMATION FOR E-COMMERCE Yutro — the business layer
AGENT RUNTIME + SHARED MEMORY Hermes, Pi, MemPalace — the machine layer
GENERATIVE MEDIA PIPELINES image / video / voice — the output layer
RAG + KNOWLEDGE INFRASTRUCTURE retrieval that has to be correct, not pretty
GEO / AEO ENGINEERING making a shop legible and citable to LLM crawlers
EDGE-NATIVE FULL-STACK Cloudflare Workers/Pages/KV/D1, Next.js, Astro
Filter I apply to every decision: does it work, does it scale sanely, does it move the number. If an elegant solution does none of those, it gets simplified until it does.
| Project | What it is | Stack | Link |
|---|---|---|---|
| Yutro | AI + UI studio for e-commerce — OpisAI, SklepGPT, ContentFactory, AI SEO | Next.js · Cloudflare | yutro.company |
| ai.yutro.company | AI store-audit platform, shop agents, SklepGPT widget | Next.js 16 · React 19 | ai.yutro.company |
| mebely.company | Headless furniture storefront over Shopify Storefront API | Next.js · Shopify | mebely.company |
| PUMO AI Knowledge Base | 2333-product catalogue made citable to LLMs; 64 AI buying guides | Astro · D1 · Vectorize | pumo-guide |
| ZENO Browser | AI browser with agent control plane, MCP tools, stateless file handoff | Astro 5 · React 18 · Electron | zen-bro-wser.org |
| BONZO ART Studio | Generative media workspace — image/video/audio, multi-provider routing, job queue | TypeScript · FAL · Replicate | BONZO_ART_studio |
| BONZO Camera Studio | Webcam avatar studio — Kling Avatar v2 lip-sync, LivePortrait | HTML · Python | BONZO_camera_studio |
| JIMBO77 | AI/automation portal, community forum, agent network | Next.js 16 · Three.js · Workers | jimbo77.org |
| DEVz HUB | Command center + architecture graph, agent monitoring | React Flow | DEVz_HUB |
One of the things I do that most shops and agencies still don't: I publish a dedicated, attributed, machine-readable channel for LLM crawlers next to the human site. Not cloaking, not fake bot targeting — a public mirror that points back to the canonical source and stays versioned.
/<channel>/llms.txt short brief for LLMs
/<channel>/llms-full.txt full brief — entities, metrics, tables, ready answers
/<channel>/llm-schema.json Schema.org @graph: Organization, WebSite, ItemList, FAQPage
/<channel>/ai-signal.json machine freshness + data-quality signal
/<channel>/aiseo-freshness.json human changelog + tip of the day
/sitemap-ai.xml AI sitemap with namespaced <ai:*> metadata
/robots.txt explicit per-agent rules: GPTBot, ChatGPT-User,
CCBot, anthropic-ai, Claude-Web, PerplexityBot
Live channels: pumo-guide · yutro · ai.yutro.company · jimbo77.org · yutro.company
Rules encoded in every channel: consistent numbers everywhere (inconsistent public numbers actively damage GEO), explicit attribution to the canonical source, versioned freshness, and no dark patterns.
The PUMO catalogue channel, measured — not estimated:
2333 active products, cleaned and verified against Allegro listings
49 categories
64 AI-generated buying guides
690 products audited for description/attribute consistency
230 mismatches found and fixed · 0 open issues
54x return on the content pipeline (self-hosted Umami analytics)
68% LLM citation rate on the channel (self-hosted Umami analytics)
Data pipeline: deterministic parser + consistency audit. Content generation: WHITECAT, a mixture-of-agents system (GPT-4 + Claude) over live catalogue data in Cloudflare D1.
LANGUAGES TypeScript · JavaScript · Python · Kotlin · SQL · Bash / PowerShell
FRONTEND Next.js (App Router) · React 19 · Astro · Three.js / R3F
Tailwind · Zustand · Framer Motion · Canvas / SVG / shaders
BACKEND Node.js · FastAPI · Python services · Express · WebSocket
EDGE Cloudflare Workers · Pages · KV · D1 · R2 · Durable Objects ·
Vectorize · Tunnel · Email Routing · Workers AI
DATA PostgreSQL · SQLite · Redis · ChromaDB · FTS5 · DuckDB
AI / LLM OpenRouter · Anthropic · OpenAI · Gemini · DeepSeek · Ollama
RAG · embeddings · fine-tuning · tool/function calling
AGENTS MCP servers · multi-agent orchestration · CLI agents
Hermes · Pi · Claude Code · Codex · Cursor · OpenCode
MEDIA SD / A1111 · ComfyUI · FAL · Replicate · FFmpeg · TTS
OPS Docker / Podman · WSL2 · git-bash · PowerShell · pnpm / bun / uv
n8n-class automation · cron · observability dashboards
READ BEFORE WRITE never guess — read the source, then edit
MINIMAL DIFF smallest change that solves the real problem
VERIFY AFTER CHANGE test / lint / smoke / sanity — no exceptions
EXPLICIT OVER MAGIC readable code beats clever code
PRODUCTION OVER EGO "works now" wins unless it is flagged as an infra bet
NO SECRETS IN OUTPUT keys, tokens, PATs — never in a repo, never in a log
SAY THE DOWNSIDE name future tech debt out loud, immediately
Interface doctrine: background #090b10, radius 0px, JetBrains Mono, technical
icons only, no emoji.
- Agent memory that is agent-agnostic — one shared palace, many runtimes
- GEO/AEO as a first-class deliverable, with freshness signals and schema
- Production retail agents measured by margin, not by demos
- Generative media at queue scale — image, video and voice in one system
- Edge-native AI: inference and routing close to the customer
STUDIO https://yutro.company
AI PLATFORM https://ai.yutro.company
BLOG / GEO https://www.mybonzoaiblog.com
AGENT CLUB https://jimbo77.org
GITHUB https://github.com/Bonzokoles
YUTRO doyutro@yahoo.com
BLOG / PUMO stolarnia.ams@gmail.com
FEAR CAUSES HESITATION. HESITATION MAKES YOUR WORST FEARS REAL.
BUILD WHAT TERRIFIES YOU.