Product-lifecycle data to the Databricks Lakehouse: ETL, MLflow pipelines, Power BI reporting.
~$1M of decisions informed / yrBerke
Kavak.
Tailor-made AI agents and RAG systems that ship and stay in production.
Independent AI & agent consulting from Berlin, for teams across the EU and US, on your own cloud and data. Production ML and GenAI since 2018.
Services
Built on your own data and your own cloud. You own the result.
Agents and RAG
Assistants that answer from your documents, and agents that act inside your tools, with evaluations so you know when they break.
Semantic search
Hybrid retrieval and reranking, so people find the right answer instead of the right keyword.
Demand and sales forecasting
Forecasts wired into planning, not left in a notebook.
Churn and retention
Who is about to leave and who is worth keeping, delivered into the CRM.
Fraud and payment audit
Suspicious claims and payment errors flagged before the money goes out.
Pricing and segmentation
Micro segmentation that gives pricing teams a sharper view than the rate table.
Text and document processing
Classification and extraction for tickets, clinical notes and forms, at volume.
Tailor made AI video
Product and brand videos generated on my own GPUs, trained on your visual style.
Selected work
Projects I have led or built, at BeeBI Consulting and TAZI AI, for clients in insurance, finance, healthcare, consumer goods and sports.
Supply planning & automation for The Quality Group's brands, plus ETL pipelines on Databricks feeding Tableau dashboards.
Implementation: months → weeks
Three ML projects end to end: sales forecasting, ideal customer retention, cross-sell / up-sell.
+1.5% retention ≈ $107k / yrClaims fraud detection and payment-audit models, deployed into the claims process.
Error detection: 1-in-8 → 1-in-4
Customer churn prediction for Türkiye's leading investment house.
At-risk clients flagged earlyAuto-insurance micro-segmentation for profitability and growth at Sompo Seguros Brasil, guiding pricing and product decisions across 500K+ policies a year.
Actuary effectiveness up to +50%MRI component fault prediction as technical lead on the EU research project, plus ITEA's medicine-cabinet project.
2 EU consortium projectsNot every client can be named. Recent projects under NDA:
- Agentic RAG chatbot, built end to end for a maritime platform startup
- Biopsy-triage model that doubled the patients in need identified
- Semantic search on AWS with hybrid retrieval and reranking
- Production NLP for telecom and food delivery brands
How we work
01
Intro call
You describe the problem. I tell you honestly whether AI is the right tool.
02
Discovery
Two weeks, a thin prototype on your data, and a fixed price for the build.
03
Build
Working software every week, demoed on your data, with evaluations from day one.
04
Run
Your team runs it. You own the code, prompts and infrastructure.
Experience
BeeBI Consulting GmbH · Berlin
Senior Data Scientist, AI Services
Running the firm's AI & GenAI portfolio for EU and US customers, from presales through post-deployment across five concurrent clients. Standardized delivery on AWS and Databricks, cutting implementation from months to weeks.
TAZI AI · Istanbul
Data Scientist · AI Solutions Manager
Early team member through 4× growth and a $4.6M round. As AI Solutions Manager, packaged TAZI's AI solutions into repeatable offerings and built the ML delivery pipeline and go-to-market; shipped retention, NLP and clinical models for US and European clients on Spark and MLflow, deployed on-premise and on SageMaker, GCP and Azure.
Earlier
Internships
Finance at Abbott Global Healthcare · IT at Vodafone, Istanbul.
About
I work at the intersection of applied research and delivery: comfortable reading the paper, but judged on whether the system runs in production and moves a business metric.
My master's research at Boğaziçi University on biomedical entity normalization with LLMs, across Spanish, English and German, was published at the BioCreative VIII SympTEMIST track. Before that, an industrial engineering and economics double major at Koç University taught me to treat ML problems as business problems first.
Based in Berlin. I work in English and Turkish, and speak good German.
- Boğaziçi University · 2020–2023
- Koç University · double major · Vehbi Koç Honor Award
- Technical University of Munich · School of Management
- Symptom normalization using unsupervised learning and text similarity · Kavak & Özgür, BioCreative VIII
- Microsoft PL-300 · Power BI Data Analyst
- Python · SQL · TypeScript · AWS · Databricks · GCP · Azure · Docker · Elastic · MLflow
Questions
Who do you work with?
Mid-sized companies and funded startups in the EU and US, usually a product or data team that needs a working system.
Which models and clouds?
Anthropic, OpenAI, Mistral or open-weight models, on AWS, Azure, GCP or Databricks. I don't resell any of them.
Fixed price or time and materials?
Scoped projects are fixed price. Ongoing work is a monthly retainer or a day rate.
Remote or on-site?
Remote from Berlin by default, on-site in Germany and Türkiye when it helps.
Can you sign an NDA first?
Yes, before the first call if you prefer.
Tell me about the problem.
Thirty minutes, no slides. Bring the business problem, I'll tell you what it would take.
berkekavak@hotmail.com