Patrick Fleith
Portrait photo

Freelance AI Engineer LLMs & Datasets

I help small AI teams with on-demand AI Engineering support, LLMs, custom dataset creation,

and AI Product R&D - without the cost and overhead of a full-time hire.

Only one client spot left
Book a call What I help with
Python (mono) Python (color)
FastAPI
Django
Hugging Face (Pirate) Hugging Face
OpenAI
PyTorch (mono) PyTorch (color)
scikit-learn (mono) scikit-learn (color)
LangChain (mono) LangChain (color)
Gradio (mono) Gradio (color)
Ollama (default) Ollama (hero)
Docker (mono) Docker (color)

Trusted by

Solenix ESA Thales Alenia Space SMT
Solenix ESA Thales Alenia Space SMT

What I Help With

Senior-level AI engineering support across LLMs projects and datasets.

Robot

LLM Engineering Setup

  • Compare and select the right LLM for your needs
  • Deploy on-premise LLM inference servers
  • Set up cloud-based LLM endpoints
  • Perform load testing
  • Prompt engineering
Smirking Face

Structured Data & Information Extraction

  • Parse raw documents like PDF, into AI-ready datasets
  • Detect and mask personally identifiable information
  • Chunking, metadata extraction, and enrichment
  • RAG datasets for vector store ingestion
Gear

Domain-Specific Dataset Engineering

  • Generate domain-specific synthetic datasets
  • Set up annotation campaigns and quality control
  • Process, clean, filter, and deduplicate datasets
  • Source, curate, and merge existing public datasets
  • Write dataset cards, guides, reports, and create visuals
Rocket

AI Product R&D

  • User interviews to identify high-impact use cases
  • Build AI assistants, RAG systems, and agents
  • Systematic evaluation to improve your product
  • Advisory for efficient AI development
  • Validate your idea shipping an MVP

How It Works

7 clear steps to get your AI project delivered

Discovery Call

A 45min call to understand your goals & needs, how I can help and ensure I am a good fit.

Onboarding

We do the paperwork, I give you access to private task board, your bonuses, and we setup a clear communication channel.

Backlog Warmup

Work begins. We create the first backlog tasks together to ensure a smooth start.

Building & Shipping

I pick ready tasks and start shipping weekly. Mostly asynchronous work for maximum focus and productivity.

More Tasks, More Commits

Create and prioritize additional tasks. We exchange and clarify them using the board until they are ready. You stay in control with full progress visibility and transparency.

Weekly Check-in

30–45 minutes, optional. For backlog refinement or complex technical topics discussion.

Monthly Highlights

Receive a summary of delivered work and impact to report to management.

Continue, pause for a few weeks, or cancel at any time

40 hrs/mo

You get 40 hours of dedicated AI/LLM & dataset engineering work every month.

Senior-level expertise

Satisfaction Guarantee

Not satisfied? Get 2 months free to complete outstanding tasks

Risk-free engagement
Exclusive Bonuses

Everything You Need for Success.

200 GPU Hours every month

RTX5090 or equivalent, unused hours add up

Europe-based Infrastructure

Inference & data storage (if needed)

HuggingFace organization setup

Manage public / private datasets, models, and demos.

Direct WhatsApp

For very urgent questions, blockers or problems

Private Task Board

Project management

Mystery Gift

Something special awaits...

Total value: €2,000+ included free

Pricing

€4,500/mo

excl. VAT

  • 40 hours of AI/LLM & dataset engineering
  • No lock-in. Cancel, pause, or resume at any time
  • Monthly highlights report
  • Weekly meeting (if needed)
Bonuses
  • 200 GPU hours (RTX5090 or equivalent) every month
  • Direct WhatsApp for urgent questions
  • Private collection and dataset viewer on HuggingFace
  • Private project task board
  • Europe-based inference & data storage if needed
  • Mystery gift
Satisfaction Guarantee

If you're not satisfied, I'll cancel your subscription and keep working for two more months for free to complete any outstanding tasks.

Book a discovery call

I serve maximum two clients in parallel to maintain high-quality work.

Only one spot left

Is this for you?

Not if you…

Require in-office presence

↳ I am already working remotely since 2020.

Expect constant and real-time availability

↳ I work asynchronously and batch communications to stay focus.

Want to choose my development tools and setup

↳ Sticking with my usual toolset compounds for speed and quality.

Prioritize process over impact

↳ I value more impact and experiments, than rigid processes.

Yes, if you…

Want flexible support + guaranteed results

↳ Cancel anytime. Get 2 months for free if not satisfied.

Need affordable senior-level expertise

↳ 7+ years of experience, no hiring headache, fraction of a full-time salary.

Prefer self-managed professionals, and don't mind async work

↳ You'll see progress while you sleep. Minimal meetings, maximum output.

Want clear, all-inclusive pricing

↳ I cover my own tools costs, GPU, incusrances; no extra costs

Next Step

Book a discovery call Send an email

FAQ

How fast can we start?
I can start within 24 hours following our discovery call.
Do you cap weekly allocation, or can you spike hours when needed?
I usually work 8 to 12 hours per week on your project.
How are the 40 hours tracked and reported?
I track my hours with Clockify and include a summary in the monthly highlights report.
Why is there only one spot "available"? Is this a trick?
No, it's true. I can only serve two clients in parallel to maintain high-quality work. I'm already in a long-term engagement with one client.
Why is it so expensive?
It is actually no that expensive, if you factor in the bonuses, the unmatched guarantee, my operating costs, and huge amount of time I spend to test new libraries, read papers, and keeping up with the state of the art in this very fast-evolving field. As a result, you won't find true experts cheaper with similar level of service.
How does cancellation work?
You can cancel anytime, even mid-month. If I've worked 20 hours, I'll refund the remaining 20.
How does pause/resume work?
If you pause, I'll stop working and you keep the remaining hours for when you resume, perfect for accommodating your timelines.
What is the difference between pause and cancel?
After a cancellation, a new customer may take the spot. If you pause, you keep priority. I'll ask you first about resuming before accepting a new client.
How does the satisfaction guarantee work?
If you're not satisfied, I'll cancel your subscription, refund unused hours, and keep working for two more months for free to complete ongoing tasks until you're fully happy. To prevent abuse and because I believe it takes at least a 40h to deliver a first batch of meaningful work, the guarantee takes effect only at the beginning of the second month. You can always cancel during the first month and I'll refund you unused hours as per the cancellation policy. However, the guarantee will not apply.
How can we collaborate efficiently if you work asynchronously?
Async work enables long, interruption-free focus blocks, reduces meetings and context switching, scales across time zones, and encourages clear written communication. You'll also see progress while you sleep or over weekends.
I don't want to pay upfront.
Advance monthly payment enables the included bonuses and the satisfaction guarantee. On a case-by-case basis, I may accept different payment terms, but then the bonuses and guarantee won't apply.
What if I don't need GPU time but need OpenAI or Anthropic API credits?
No problem, I'll convert the included €200 into OpenAI or Anthropic API credits.
Can we purchase additional hours? What are overage rates?
If you need more and if I am available, you can purchase additional hours. I'll invoice at €80/hour.
What if we need more GPU or API credits?
Additional costs will be invoiced at the end of the month. The included 200 GPU hours are a reasonable starting point and cover most typical use cases.
What does "Europe-based infrastructure (inference and storage)" mean?
If your policy requires it, I can set up inference endpoints and data storage with a European cloud provider (useful for data residency and compliance).
Can I access the GPU myself?
The offered GPU hours are for work I perform on your project. For security reasons, clients can't access my AI workstation.
What's your vacation/holiday policy and coverage during absences?
I'll give at least 10 days' notice before vacations. Your subscription will automatically pause during absences and resume afterward. If you can't wait, I'll arrange a replacement.
What happens if hours are unused? Do they roll over?
No, unused hours do not roll over. It is preferrable to pause the subscription is you don't have enough work.
What tools do you use for the "private task board"?
By default I use a private GitHub Project. I can also set it up in your GitLab or Jira if you prefer.
Can you work under our Git, cloud accounts, and security policies?
Yes. I've worked within several clients' infrastructures where security is critical (e.g., space agencies).
Where are you based?
My company is based in France.
What language(s) do you work in?
I can fluently work and communicate in English and French (native).
What's included for post-delivery support and bug fixes?
If there is a bug, we can create a new task in our shared task board, and depending on its priority I can pick it up to work on it.
What exactly is included in the "Hugging Face organization setup"?
Hugging Face is like GitHub for AI models and datasets. You can showcase work publicly or use it privately for internal collaboration. With this workspace you'll be able to store models, datasets, host demos, and have access to a private dataset viewer.
Can you help recruit/hand off to a full-time hire later?
Yes. I can support the recruitment process and hand off to a full-time hire, though recruitment isn't my core expertise.