# PUSUCIP | AI Studio

> AI that works, for real businesses. No buzzwords — just AI implementation.

## About
- **Founder:** Pusu Cip (Ciprian Cosmin Muresan)
- **Location:** Turda, Cluj, Romania
- **Contact:** pusucip25@gmail.com | +40 765 121 976
- **Languages:** Romanian, English
- **Site:** https://pusucip.uk

A solo AI automation studio. Everything is built and delivered by one person, which is the point: no account managers, no handover loss.

## Services
- **AI Video Production** — script to screen. Short-form series, promotional reels, talking avatars.
- **Websites & Automation (n8n)** — business sites plus n8n workflows connecting email, WhatsApp and APIs.
- **AI Consulting** — finding where AI actually fits a business, then building it.
- **Content & Marketing** — AI-assisted content for social, email and SEO, edited by hand.
- **Export & Trade** — connecting Romanian artisan producers with EU markets: research, buyer targeting, export pipeline.
- **Agent Systems (Hermes)** — multi-agent workflows with persistent memory, subagents and scheduling.

## Projects
- [NodeQuest](https://nodequest.io) — browser puzzle game that teaches workflow automation. 38 levels (28 story + 10 PRO), real n8n JSON export. Live.
- AiCeo.ro — AI CEO platform. Building (not public yet).
- Dentistiq — video content production. Live.
- Cheese Export — Italy pipeline for Romanian producers. Live.
- TikTok AI Drama Series — automated short-form content. Live.
- Pinterest Bot v2 — affiliate automation. Building.
- Hermes Agent Ecosystem — multi-agent orchestration. Active.
- Transport Brokerage — voice AI. Planning.

## Working with the studio
- Projects are quoted individually in writing, with scope and price.
- Work uses n8n for orchestration, Cloudflare for hosting, and a mix of open and commercial AI models depending on the task.
- Typical engagements: a single automation, a content pipeline, or an end-to-end system with an agent layer.

## Verification
Work is checked before it ships, not eyeballed. Content and decisions are reviewed with **Jev** (TypeSafe's decision model) on questions where a wrong answer is expensive.

Measured example — all 38 NodeQuest levels were rated *solvable without guessing* by Jev, mean 0.79, none below 0.5. A control run with the hints stripped scored 0.46 and pushed 26 of 38 under the threshold; that gap is what makes the number meaningful rather than an automatic approval. Full table: <https://nodequest.io/llms-full.txt>.

## Pages
- [Home](https://pusucip.uk/) — services and portfolio
- [Privacy policy](https://pusucip.uk/privacy)
- [Terms of use](https://pusucip.uk/terms)

## Keywords
AI automation studio, n8n automation, AI video production, AI consulting, workflow automation, multi-agent systems, Romania, Cluj, Turda, EU export automation