Paris-based · EU-hosted · taking new projects

From AI demo
to production.

By people who actually ship.

A Paris-based team of senior SWE / MLE engineers building Claude Code agents that survive contact with real users — not another consulting deck.

YC TECH · PSL « AI & Management » · multi-industry operator

Talk to us → See what we ship
Or self-serve — 5 seconds to start
~/your-project 
$ export ANTHROPIC_BASE_URL=https://api.youchun.tech
$ export ANTHROPIC_API_KEY=sk-your-key
$ claude
// The problem

Everyone builds demos.
Almost no one ships.

01

Six months of R&D, zero in production

The agent looks great in the all-hands. It melts under real load.
The board claps. Zero customers ever touched it.
You know AI is the future — you just can't find anyone who can ship it inside your company.

02

Hiring AI engineers is broken

Every CV since 2023 says "LLM experience".
Almost none have run an agent in production for six months without incident.
You don't need a consultant who PowerPoints. You need someone who's been on-call for what they built.

03

$200 / dev / month adds up fast

Claude Max for a 10-person team: $24K / year on API alone.
Plus everyone rebuilds the same integration plumbing in parallel.
You want one place that handles infra, billing, and onboarding — so your devs build, not configure.

// Services

Four ways we put AI to work in your company.

No slides. We get into your code, understand your business, ship something that runs.

01

AI workflow design + end-to-end delivery

From process audit to agent in production. We map your business flow, find where AI actually moves a number (not where it just looks cool), build, deploy, monitor.

02

Private agents · RAG · multi-agent systems

Deployed on your infra (AWS, GCP, Azure, on-prem). Your data stays where it is. EU hosting available. Stack: Claude Code, LangGraph, your choice of vector DB.

03

AI uplift on existing operations

You're already running — support, content, document processing, internal ops. We graft AI where it creates measurable value, not where it makes a good keynote.

04

Knowledge transfer to your in-house team

We don't want to be your vendor forever. By the time we leave, your team can maintain and extend what we built. Docs, runbooks, live training, code commented in your team's working language.

We're not an AI agency. We're operators who built our own businesses with these tools — and now make the same expertise available to yours.

// Team

We use it on ourselves first.
Then we ship it to you.

YC TECH
name YC TECH
based Paris · FR
role founder · operator
github xiao98 ↗
email xiaohaofr@gmail.com

The founder — student at PSL « AI & Management » — runs 5 live businesses built on Claude Code.

Not hackathon projects. Real customers, real invoices, real production incidents. What we ship to you, we've already shipped to ourselves.

For B2B engagements, we mobilize a network of senior SWE / MLE engineers (10+ years across big tech and scale-ups), assembled per project to match your scope.

// Profile
  • ›PSL « AI & Management », currently enrolled
  • ›6 ventures launched (edtech, SaaS, content, AI infra)
  • ›Background in data engineering & business consulting
  • ›Asia–Europe bridge: 5,000+ contacts in business / academia / tech
  • ›Trilingual: English · French · Mandarin
// Cases

5 products we currently run.

All built on Claude Code. All in production. All with users who pay or rely on them daily.

01
Facture-Agent
SaaS for restaurant invoice OCR + automated VAT filing in France
Claude Vision · FastAPI · Vue3 · FR market
02
AI content production line
End-to-end pipeline: research → drafting → multi-platform publishing
Multi-agent · LinkedIn · WeChat · Video
03
Automated competitive intel
Scraping + transcription + analysis — 300+ videos already ingested
MediaCrawler · Whisper · Lark
04
Personal quant research lab
Multi-strategy crypto backtests (Wyckoff + DTW + funding rate)
Python · pandas · TA-Lib · CCXT
05
Atlas Think Tank — research report production line
Geopolitical & cultural analyses delivered at institutional standard (BlackRock BII style)
Multi-source research · synthesis · PDF / slide generation
// Claude lineup

Opus 4.7 available now

Full Claude Code lineup. Not throttled, not downgraded, no added latency.

LATEST
claude-opus-4-7 · 1M context
Opus 4.7
Deep reasoning · long-horizon agents · large codebase comprehension
claude-sonnet-4-6
Sonnet 4.6
Daily workhorse · balanced speed and quality
claude-haiku-4-5
Haiku 4.5
Lowest latency · batch jobs · cost-efficient
// Pricing

Two ways to work with us.

RECOMMENDED
B2B · Custom engagement
Custom quote
  • ✓Free 1-hour scoping call, no commitment
  • ✓Senior SWE / MLE team assembled per project
  • ✓End-to-end delivery: design → production → handoff
  • ✓Fixed-fee or T&M, whichever fits your procurement
  • ✓EU-hosted option, GDPR-friendly, DPA on request
Request a quote
Self-serve · Claude Code
$80 / month
  • ✓Full Claude Code access (Opus 4.7 / Sonnet 4.6 / Haiku 4.5)
  • ✓Claude Max-tier quota
  • ✓1M context, no degradation
  • ✓7-day full refund
Create account

Not sure what you need? Email us. We answer in 24h with an honest take — sometimes that means "you don't actually need us yet."

Let's talk.

30 minutes, no commitment. Describe the problem, we tell you honestly whether we can help.

xiaohaofr@gmail.com

Paris-based · reply within 24h on business days

// FAQ

Common questions

Are you an agency or freelancers? +
Neither. We're operators — we run our own businesses on Claude Code, and we open part of our time to B2B engagements where we can add real value. Each engagement is staffed with a small team of senior engineers matched to scope.
How do you bill? +
Fixed-fee for scoped engagements (best for annual budgets). Time & materials for exploratory work. We invoice in EUR or USD; net-30 standard. Capable of issuing French-compliant invoices for EU clients.
What stacks do you work with? +
Backend: Python (FastAPI, LangGraph), Node, Go. Frontend: React, Next, Vue 3. Hosting: AWS, GCP, Scaleway, on-prem. Models: Claude (preferred), GPT, open-weight models when sovereignty requires it. We pick what fits your context — no dogma.
Data privacy and EU compliance? +
EU hosting (Scaleway Paris, OVH) available. DPA on request. For sensitive data (health, HR, finance), we deploy fully on-prem or to your private cloud — Claude Code stays usable in self-hosted gateway mode.
How long for a working POC? +
2–4 weeks for a functional POC on a scoped use case. 6–12 weeks to a stable production system. We'd rather ship small and early than aim for the moon and miss the runway.
Do you work in time zones outside Europe? +
Yes. The team is distributed; we cover EU and Asian time zones natively, US working hours via overlap windows. Async-first — daily written updates, sync calls only when needed.
What's the $80/mo self-serve plan exactly? +
Claude Code access (Opus 4.7 / Sonnet 4.6 / Haiku 4.5) at Claude Max-tier quota. Good for solo devs, small teams, or as a sandbox before a B2B engagement. Full refund within 7 days if it's not for you.