Kaan Goker · Leuven, Belgium

Machine Learning / AI Engineer & Data Scientist

I build end-to-end AI systems in Python: NLP, LLMs, RAG, and speech, deployed with FastAPI and Docker. Advanced Master of AI, KU Leuven.

Open to AI / ML / Data Scientist roles in Belgium, hybrid or remote. Available now. Download the CV (PDF).
See projects Get in touch
100%
Job Success on Upwork
2
Live demos on Hugging Face
27
QLoRA fine-tuning runs
99
Languages, one subtitle pipeline
// featured projects

Systems that run, not slideware

Every card is the same story: a real problem, the system I built for it, and a result you can check. Several of them are live right now.

Data Pipelines & LLM Extraction

Belgian AI Jobs Radar

Problem: nobody knows what Belgian employers actually hire for in AI, week by week. System: a daily pipeline that ingests every new AI job posting in Belgium, deduplicates it, extracts structured signals with an LLM (role, stack, seniority, language demands, work mode) into SQLite, and publishes an interactive dashboard. It runs unattended on a GitHub Actions cron with a regression guard on every export.

400+ active AI roles tracked, refreshed every day
PythonGeminiSQLiteEChartsGitHub ActionsAzure
GitHub Live dashboard: this week
EU AI Act RAG Screener interface
RAG & LLM Systems

EU AI Act RAG Screener

Problem: teams building AI products can't easily tell where their system lands under the EU AI Act. System: a RAG chatbot that classifies risk level and backs every answer with page-level citations from the official EUR-Lex text, so nothing is hallucinated legal advice.

Live demo, every claim traceable to its source page
GeminiLangChainChromaDBChainlitDocker
Live demo GitHub
GetSubtitlesApp interface
Speech & Media AI

GetSubtitlesApp

Problem: subtitling video by hand is slow, and most tools cap the language list. System: a desktop and web app that turns any audio or video into .srt subtitles with Whisper, with translate-to-English and vertical-video modes, shipped as a Docker image and a Windows .exe.

99 languages, live on Hugging Face Spaces, 200+ downloads
faster-whisperFastAPIStreamlitDockerPyInstaller
Live demo GitHub
Latin poetry generation pipeline diagram
NLP & Fine-tuning

Latin Poetry Generator

M.Sc. thesis. LLMs can't hold strict formal constraints like Latin meter on their own, so I paired QLoRA fine-tuned Llama and Gemma models with a rule-based metrical validator in an iterative feedback loop, on a dataset I built and published.

27 fine-tuning runs, dataset published on Hugging Face
PyTorchQLoRA/PEFTHugging FaceCLTKW&B
GitHub Dataset
Data Science

Belgium Tech Labour Pulse

Is Belgium a good market for tech talent? 22 years of Eurostat data answer it: a paired t-test with bootstrap confidence intervals (Belgium grows at the EU pace, p = 0.64, and the report says so), k-means clustering of countries, and a backtested forecast to 2027, all in one reproducible pipeline with 15 tests.

Honest statistics, single-file interactive dashboard
pandasscipyscikit-learnstatsmodelsECharts
GitHub
Voice AI

Receptionist

A trilingual (Dutch, French, English) voice receptionist for small Belgian businesses: it listens, answers in the caller's language, checks availability and books appointments into SQLite through LLM function calling. Business facts live in one editable persona file, so the same system serves a different shop in minutes.

Speech in and out, bookings confirmed in three languages
FastAPIGemini function callingWeb Speech APISQLite
GitHub Live demo: this week
RAG & LLM Systems

DocAgent

Upload any PDF and ask questions: every answer carries [page N] citations, and when the document does not contain the answer the agent says so instead of guessing. Export a brief or fire a webhook with the findings.

Cites or refuses, never invents
GradioChromaDBGeminiPython
Live demo GitHub
Civilization VII Turkish localization mod preview
Agents & Automation

Civ VII Turkish Localization

Civilization VII has no official Turkish, so I built a 6-step pipeline that extracts 43,000+ game strings, builds a 496-term glossary for consistency, routes long narrative text to a strong model and short UI strings to a cheap one in parallel, and validates every placeholder token before building the mod.

Published on Steam Workshop, 150+ subscribers
PythonClaude CLIGemini CLIModel routingParallelism
Shipped as a Steam Workshop mod
Art app home screen
Multimodal, side project

Art: vision-LLM mobile app

Point your camera at any image and the app scores which famous artists' styles it echoes, with structured reasoning behind each score. Gemini vision does the analysis through a 4-step prompt; a self-hosted FastAPI + CLIP backend gives an offline fallback, and a Cloudflare Worker keeps the keys off the device.

Bilingual (EN/TR), camera + draw-and-analyze canvas
React NativeGemini VisionCLIPFastAPICloudflare
In development
// stack

What I work with

AI / ML

NLPASR / SpeechRAGLLMsPyTorchHugging FaceQLoRA / PEFTLangChainscikit-learnPrompt engineeringAgentic workflows

Engineering

PythonFastAPIDockerREST APIsSQLGitLinuxBash

Data

PandasNumPyMatplotlibETL pipelinesKPI dashboardsData cleaning

Cloud & Tools

Google CloudStreamlitChainlitWeights & BiasesChromaDBQdrantpgvectorn8n / MakeFFmpeg
// experience

Where the systems ran

Dec 2025 - present

Freelance AI Engineer & Data Analyst

Upwork · Remote
  • Build AI agents, agentic workflows and automation systems that put LLMs, RAG and Voice AI into client applications.
  • Deliver data analysis: predictive models, KPI dashboards and clear recommendations (Python, Pandas, scikit-learn, SQL).
  • 100% Job Success, five stars on every contract so far.
Oct 2022 - Sep 2024

Digital Media Analyst

Turkishle · Remote (USA)
  • Built a Python pipeline (Whisper + FFmpeg) that generated, split and formatted subtitles across 99 languages.
  • Built a Python ETL feeding channel analytics into weekly KPI dashboards; the tests it guided on titles and intros contributed to a ~15% watch-time increase.
  • Managed the LMS and online store; automated scheduled email campaigns for launches.
Sep 2020 - Jun 2023

Digital Content Strategist

KU Leuven, Pangaea · Leuven
  • Designed data-driven content strategies and optimized engagement using platform analytics; organized university-wide multicultural events.
Jan 2022 - Sep 2022

Editorial Coordinator & Digital Content Producer

FOMU, Fotomuseum Antwerpen · Antwerp
  • Managed the CMS for online and print publishing; editorial coordination, image licensing and content curation.
// education

KU Leuven, twice over

2024 - 2026

Advanced Master of Artificial Intelligence, KU Leuven

Specialization in NLP and Automatic Speech Recognition. Thesis: hybrid LLM fine-tuning with a symbolic feedback loop for constrained Latin poetry generation. Coursework spans machine learning, NLP, speech recognition, language engineering and AI ethics.

2021 - 2023

MA Cultural Studies, KU Leuven

Digital media, ethics and language.

2013 - 2020

Philosophy (BA + MA), Yeditepe University & KU Leuven

Logic, cognition and the mind-body problem. The rigor came from here.

// about

From philosophy to production

I started in philosophy, spent years on logic and how minds work, then moved through digital media into AI. The Advanced Master at KU Leuven made it technical. My thesis ties the whole path together: it combines fine-tuned LLMs with symbolic logic to generate poetry that has to follow strict formal rules.

That background shapes how I build. I care about systems that can explain themselves: RAG answers with real citations, generation checked by validators, pipelines that log what they did. I like shipping the whole thing, from the dataset to the Docker container.

Based in Leuven since 2017. When I'm not training models I'm probably on a tennis court or editing video, which is how I got into speech AI in the first place.

:: languages

Englishprofessional
Turkishnative
DutchA2
Frenchelementary
// contact

Let's build something

Looking for an AI / ML engineer or data scientist in Belgium, hybrid or remote? I answer fast.

nadirkaangoker@gmail.com LinkedIn