About — 678 Studio

I'm Ali. An independent data scientist working through 678 Studio.

Data Scientist & AI Developer · South Holland, the Netherlands · MSc Data Science, Tilburg University.

Based in: South Holland
Studio: 678 Studio
Education: MSc, Tilburg
Languages: EN · NL · FA · SV

Bio

“My job is to make AI feel less like a gamble and more like a good decision.”

I'm Ali — the person behind 678 Studio. I grew up curious about how things work and ended up building a career out of turning messy data into decisions people can actually act on. Today I help ambitious teams in the Netherlands and across the EU put AI to work, without the hype.

Working with me means working with one senior person end-to-end. No account managers, no handoffs, no juniors learning on your budget. I sit with your team, understand the business, and ship things that hold up in production — AI agents, LLM assistants and RAG systems, forecasting and predictive models, dashboards, data pipelines and evaluation frameworks.

Clients hire me when they want a trusted partner to translate an idea into a working product, rescue a stalled AI project, or bring rigour to something that's been running on gut feel. I bring an academic foundation in data science, years of hands-on engineering, and a bias toward measurable outcomes over slideware.

What I can do for you

A calm pair of hands for the hard AI questions.

Most clients come to me with one of four problems. Here is how I usually help — quickly, quietly, and with something real to show at the end.

Turn an idea into a working AI product

You have a concept — an assistant, a smarter search, a recommendation — and need someone who can take it from whiteboard to a live product your users can actually use.

Rescue or level up a stalled project

The prototype impressed everyone in the demo but never made it to production. I come in, find what is really blocking it, and get it shipped with the tests and monitoring it should have had from day one.

Bring rigour to decisions made on gut feel

Pricing, churn, forecasting, risk — the places where a good model quietly pays for itself many times over. I build ones you can trust, explain to stakeholders, and keep honest over time.

Be your fractional data & AI lead

For teams that need senior direction without a full-time hire. I set the technical strategy, work alongside your engineers, and stay involved as long as you need me — no more, no less.

Skills at a glance

The toolbox behind the work.

A quick sense of what I reach for, grouped by where it lives in the stack. You do not need to know any of these to work with me — picking the right tool is my job.

AI agents & GenAI

LangGraph agents
Tool-use & function calling
Multi-step orchestration
RAG pipelines
LangChain · LlamaIndex
OpenAI · Anthropic · open models
Structured outputs (Pydantic)

Evaluation & fine-tuning

RAGAS
DeepEval
PromptFoo
LoRA / QLoRA
Hugging Face · PEFT
ROUGE / BLEU

ML, data science & analytics

Python · PyTorch
scikit-learn
XGBoost · LightGBM · CatBoost
Causal inference (DiD, PSM)
Time-series & forecasting
SQL · pandas · dbt
Data pipelines & ETL
Dashboards & BI (Power BI, Tableau)

Production & MLOps

FastAPI · Docker
MLflow
Azure ML · AWS SageMaker
Drift detection
CI/CD for ML
Vector DBs (Chroma, FAISS)

How I work

Three principles
that show up in every project.

Production over notebooks

The gap between a working model and a running system is where most projects die. I ship with MLflow registries, drift monitoring, FastAPI + Docker deploys, and audit-friendly agent graphs.

Evaluation before launch

Every LLM system I ship carries a real evaluation harness — RAGAS, DeepEval, PromptFoo, ROUGE — so quality, cost and latency are numbers, not vibes.

Statistical honesty

Hypothesis tests before modelling, causal methods (DiD, propensity matching) when observational data lies, and metrics chosen for the domain — AUC/KS for credit risk, recall for medical imaging, MAE for forecasting.

Experience timeline

2026

678 Studio — Independent

Applied AI and data projects for clients across the Netherlands.

2025

Digital Society School (HvA)

Built HvA's first AI discovery platform end-to-end: RAG pipeline, RAGAS/DeepEval/PromptFoo evaluation harness, stakeholder validation and production launch.

2024

Independent portfolio

LangGraph loan-review agent, LangChain contract intelligence, LoRA fine-tuning of Qwen2.5-1.5B, RAG chatbot on ChromaDB + Ollama.