[{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/ai-ethics/","section":"Tags","summary":"","title":"Ai Ethics","type":"tags"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/categories/ai-strategy/","section":"Categories","summary":"","title":"AI Strategy","type":"categories"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/alignment/","section":"Tags","summary":"","title":"Alignment","type":"tags"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/posts/","section":"Blog","summary":"","title":"Blog","type":"posts"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/categories/","section":"Categories","summary":"","title":"Categories","type":"categories"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/","section":"Demir Technology","summary":"","title":"Demir Technology","type":"page"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/eu-ai-act/","section":"Tags","summary":"","title":"Eu Ai Act","type":"tags"},{"content":"In January 2026, OpenAI quietly shipped what they\u0026rsquo;d promised never to do: ads inside ChatGPT. Sam Altman had been on record for years saying ads would compromise the product. Then infrastructure costs caught up with him.\nFive months later, the picture is clearer than any single launch announcement could make it. This isn\u0026rsquo;t a monetisation tweak. It\u0026rsquo;s the alignment problem we were warned about, arriving on schedule — and it has direct consequences for any European company building production workloads on top of closed AI vendors.\nThe objective function flipped # The technical claim being made is that ads will live in a \u0026ldquo;separate area\u0026rdquo; and \u0026ldquo;responses will not be manipulated.\u0026rdquo; Take that at face value for a moment and the deeper problem is still there.\nUntil January, ChatGPT\u0026rsquo;s optimisation target was helpfulness — a soft, generous loss function that tried to align the model with whatever the user actually needed. With an ad layer attached to the same surface, a second loss function shows up next to the first: conversion.\nThe two are not friendly. They share user attention, screen real estate, conversational momentum, and, crucially, the inferred profile the system has built about you. The instant the model is rewarded for outcomes other than helping you, alignment with your interests starts to leak — even if no single response is technically \u0026ldquo;manipulated.\u0026rdquo;\nThis is the structural concern, not a vibe. When the trainee learns that one of the metrics that matters is whether you click, the trainee learns to nudge.\nWhy your privacy settings don\u0026rsquo;t save you # The standard counter is \u0026ldquo;users can turn ad personalisation off.\u0026rdquo; That is the classic banner-ad mental model, and it doesn\u0026rsquo;t hold for conversational AI.\nWhen you talk to ChatGPT, you don\u0026rsquo;t submit search queries; you submit unstructured, deeply contextual confessions. A worried late-night question about a side effect, a half-formed business plan you\u0026rsquo;d never put in a doc, the way you actually argue with your co-founder. None of that data is \u0026ldquo;data\u0026rdquo; in the legacy sense — it\u0026rsquo;s inferred data, the kind no opt-in checkbox covers.\nEven with personalisation disabled, the model can infer your situation from how you write, the entities you mention, the way you reason. That signal will not stay locked away from a system that has a conversion target attached.\nThe Center for Humane Technology has a phrase for this: \u0026ldquo;hacking the human mind.\u0026rdquo; It is more apt for AI assistants than for any social feed that came before.\nTransparency was already trending the wrong way # Stanford\u0026rsquo;s 2025 Foundation Model Transparency Index dropped OpenAI\u0026rsquo;s score by 14 points, moving them from second to sixth. Their training data, model internals, and post-deployment system cards have been getting steadily less open, not more. \u0026ldquo;Open\u0026rdquo; in the name is now a historical artefact.\nAgainst that backdrop, \u0026ldquo;trust us — ads won\u0026rsquo;t bias the answers\u0026rdquo; is not a serious assurance. It\u0026rsquo;s a promise with no enforcement mechanism, no audit trail, and a vendor whose financial incentives now point the other way.\nThe EU AI Act kicks in with sharper teeth in August 2026. Until then, the protection users in Europe actually have is whatever they build themselves into their architecture.\nWhat this means for European data and AI teams # If you\u0026rsquo;re a CTO, Head of Data, or AI lead at a European company, the practical takeaway is uncomfortable but simple: every critical workflow you have running through a closed US vendor is now exposed to a different incentive structure than the one it was procured under.\nThree things we\u0026rsquo;re advising clients to do right now:\nInventory your dependencies on closed model APIs. Anywhere a system prompt, a retrieved document, or a user query crosses into a third-party LLM, mark it. That is the attack surface. Move sensitive workloads to open-weight models on infrastructure you control. Llama, Mistral, Qwen, and the latest open releases are good enough for the vast majority of enterprise use cases — RAG, classification, extraction, internal copilots — when paired with a serious retrieval layer. Treat AI vendor selection as procurement, not magic. Demand transparency reports. Read the system cards. Track the FMTI. Build switching cost into the architecture, not into the contract. The companies that took data residency seriously a decade ago are the ones who weren\u0026rsquo;t scrambling when GDPR went enforcement-hard. The same will be true for AI sovereignty in 2027.\nWhere we stand # If AI breaks, everything produced through it inherits that breakage — ethically, legally, and competitively. Europe needs to take open source seriously, fund local AI infrastructure, and inform its citizens. None of that happens by accident.\nAt Demir Technology we work with European companies that want production AI without surrendering the trust layer. That usually means a mix: Microsoft Fabric for the data estate, knowledge-graph-instructed retrieval for the semantic layer, and open-weight models running on customer-controlled infrastructure for anything sensitive. The architecture is real today, not five years out.\nIf your team is rethinking its AI vendor exposure, get in touch — the first conversation is free.\n","date":"16 May 2026","externalUrl":null,"permalink":"/posts/chatgpt-ads-alignment-problem/","section":"Blog","summary":"","title":"Five months in: ChatGPT's ad model is the alignment problem we were warned about","type":"posts"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/local-llm/","section":"Tags","summary":"","title":"Local Llm","type":"tags"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/open-source-ai/","section":"Tags","summary":"","title":"Open Source Ai","type":"tags"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/openai/","section":"Tags","summary":"","title":"Openai","type":"tags"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/sovereign-ai/","section":"Tags","summary":"","title":"Sovereign Ai","type":"tags"},{"content":"","date":"16 May 2026","externalUrl":null,"permalink":"/tags/","section":"Tags","summary":"","title":"Tags","type":"tags"},{"content":"","date":"1 May 2025","externalUrl":null,"permalink":"/tags/airflow/","section":"Tags","summary":"","title":"Airflow","type":"tags"},{"content":"","date":"1 May 2025","externalUrl":null,"permalink":"/categories/data-engineering/","section":"Categories","summary":"","title":"Data Engineering","type":"categories"},{"content":"","date":"1 May 2025","externalUrl":null,"permalink":"/tags/data-engineering/","section":"Tags","summary":"","title":"Data Engineering","type":"tags"},{"content":"","date":"1 May 2025","externalUrl":null,"permalink":"/tags/dbt/","section":"Tags","summary":"","title":"Dbt","type":"tags"},{"content":"","date":"1 May 2025","externalUrl":null,"permalink":"/tags/modern-data-stack/","section":"Tags","summary":"","title":"Modern Data Stack","type":"tags"},{"content":"","date":"1 May 2025","externalUrl":null,"permalink":"/tags/snowflake/","section":"Tags","summary":"","title":"Snowflake","type":"tags"},{"content":"The Modern Data Stack has matured significantly. After several years of explosive tooling growth, the dust is settling and we can see clearly which tools have earned their place in production and which were overhyped.\nHere\u0026rsquo;s my current take after running these tools in real engagements.\nWhat\u0026rsquo;s Actually Working # dbt — Still the Foundation # dbt has won the transformation layer. The combination of SQL templating, testing, documentation, and lineage in a single framework is simply unbeatable for most use cases. Every data platform I build today has dbt at the center.\nThe critical workflow: raw → staging → intermediate → mart. Simple, traceable, testable.\nAirflow vs Prefect — It Depends # Airflow remains the enterprise standard. The ecosystem is massive, the community is huge, and managed offerings (Astronomer, MWAA, Cloud Composer) have removed much of the operational pain.\nPrefect is the better developer experience. If you\u0026rsquo;re a smaller team or starting fresh, Prefect 2.x is genuinely enjoyable to work with. Dynamic task mapping and the Python-native API are a real step forward.\nSnowflake vs Databricks — Choose Your Battle # Snowflake if you\u0026rsquo;re primarily doing SQL analytics and BI. The separation of compute and storage, instant scaling, and zero-ops model are hard to beat.\nDatabricks if you need a unified platform for both data engineering and ML. The Delta Lake format, Unity Catalog, and tight Python/Spark integration make it the better choice for ML-heavy workloads.\nWhat I\u0026rsquo;m Watching Closely # Apache Iceberg is winning the open table format war. Snowflake, Databricks, and every major cloud provider are embracing it. If you\u0026rsquo;re building a new lakehouse, build on Iceberg.\nDuckDB for local development and smaller datasets is surprisingly powerful. It\u0026rsquo;s changed how I prototype pipelines.\nWhat\u0026rsquo;s Overhyped # Real-time everything. Most businesses genuinely don\u0026rsquo;t need sub-second latency. A well-designed batch pipeline refreshing every 15 minutes solves 90% of analytics use cases at a fraction of the complexity and cost.\nWant to discuss how to build a data platform that\u0026rsquo;s right for your team\u0026rsquo;s needs and budget? Get in touch.\n","date":"1 May 2025","externalUrl":null,"permalink":"/posts/modern-data-stack-2025/","section":"Blog","summary":"","title":"The Modern Data Stack in 2025: What's Actually Worth Using","type":"posts"},{"content":"","date":"4 August 2024","externalUrl":null,"permalink":"/tags/azure/","section":"Tags","summary":"","title":"Azure","type":"tags"},{"content":"","date":"4 August 2024","externalUrl":null,"permalink":"/categories/community/","section":"Categories","summary":"","title":"Community","type":"categories"},{"content":"","date":"4 August 2024","externalUrl":null,"permalink":"/tags/community/","section":"Tags","summary":"","title":"Community","type":"tags"},{"content":"","date":"4 August 2024","externalUrl":null,"permalink":"/tags/dp-203/","section":"Tags","summary":"","title":"Dp-203","type":"tags"},{"content":"","date":"4 August 2024","externalUrl":null,"permalink":"/tags/microsoft/","section":"Tags","summary":"","title":"Microsoft","type":"tags"},{"content":"","date":"4 August 2024","externalUrl":null,"permalink":"/tags/teaching/","section":"Tags","summary":"","title":"Teaching","type":"tags"},{"content":"Hey data family! I want to share the story of a volunteer DP-203 training I ran over the past few weeks — and why it became one of the most rewarding experiences of my career.\nThe Team # Together with a team of six — Alper Kaan Kavılı, Mehmet Koç, Adem Doğan, Emrullah Çelik, Sümeyra Zenkin, and Enes Yılmaz — we embarked on an unforgettable adventure in the world of Azure Data Engineering. Every Saturday evening for two hours, we pushed through the material together.\nWhy DP-203? # The Microsoft DP-203: Data Engineering on Azure certification is one of the most direct keys to breaking into professional data engineering. It covers the full spectrum of what a modern Azure data engineer needs to know — and it does so with real, production-relevant content.\nHow We Ran It # As a Microsoft Certified Trainer (MCT), I used Microsoft\u0026rsquo;s official DP-203 courseware to give students the most up-to-date and comprehensive content available. Each week we tackled a new topic, building progressively toward a complete picture of the Azure data engineering landscape:\nAzure Data Factory — orchestration and data movement at scale Azure Databricks — distributed processing with Spark Azure Synapse Analytics — unified analytics across data warehouse and data lake Every session combined theory with hands-on lab work based on real-world scenarios. No slides-only sessions — students had to get their hands dirty in Azure every week.\nMicrosoft\u0026rsquo;s $200 Azure Credits # One of the practical highlights: Microsoft provided each student with a $200 free Azure credit, giving them a real cloud environment to experiment in without any financial barrier. This made the labs meaningful — students weren\u0026rsquo;t simulating, they were actually building.\nThe LinkedIn Community # We also created a dedicated LinkedIn group where students stayed in contact with each other and with other professionals in the industry. Each week, members shared what they\u0026rsquo;d learned, discussed their lab work, and followed the latest developments in data engineering. That ongoing conversation between sessions turned out to be as valuable as the training itself.\nWhat I Took Away # Teaching is the fastest way to deepen your own understanding. Running this training reinforced concepts I use daily and pushed me to articulate things I usually just do without thinking. Watching six people go from zero to confidently navigating Azure Data Factory pipelines in a few weeks is genuinely motivating.\nIf you\u0026rsquo;re working in data and looking to validate your Azure skills — DP-203 is worth pursuing. And if you\u0026rsquo;re an MCT or experienced practitioner: consider giving back. The data community grows stronger when we teach each other.\nInterested in Azure Data Engineering training or want to discuss a project? Get in touch.\n","date":"4 August 2024","externalUrl":null,"permalink":"/posts/dp203-volunteer-training/","section":"Blog","summary":"","title":"Volunteering as an MCT: Teaching DP-203 Azure Data Engineering","type":"posts"},{"content":" At a glance # Demir Technology is an Amsterdam-based data and AI consultancy. We operate at the intersection of Data Engineering, AI Engineering, and Data Science — building modern data estates on Microsoft Fabric and ontology-instructed AI systems that combine knowledge graphs with LLMs.\nThe work is delivered by Fatih Demir — Data Engineer, AI Specialist, and Microsoft Certified Trainer (2008–2024) with 25+ years across meteorology, cyber security, academia, and the modern data stack.\nBased in the Netherlands, working with clients across Europe — remote and on-site.\nWhat sets us apart: We go beyond standard AI by designing systems that understand business logic — integrating knowledge graphs and ontologies with large language models. Engineering rigor backed by certified Microsoft expertise. Core focus # Data Engineering \u0026amp; Microsoft Fabric. Expert in architecting unified data estates. Leveraging Microsoft Fabric together with Apache Kafka and Flink to build high-throughput ETL/ELT pipelines that ensure real-time data availability and governance.\nAI \u0026amp; GenAI Engineering. Moving beyond standard models to design ontology-instructed AI systems — integrating knowledge graphs with LLMs to create semantic engines that understand business logic, not just process text.\nData Science. Applying rigorous statistical methods and machine learning to optimization problems, validated by Azure Data Scientist credentials.\nTechnical skills # Data Engineering # Microsoft Fabric Apache Spark Apache Kafka Apache Flink dbt Apache Airflow AI \u0026amp; Semantic Systems # Python PyTorch scikit-learn LangChain OpenAI API Hugging Face Neo4j Cypher Cloud \u0026amp; Infrastructure # Azure AWS Google Cloud Snowflake Databricks Docker Kubernetes Terraform Data Foundations # PostgreSQL Delta Lake Apache Iceberg SQL Current engagements # Chief Innovation Officer \u0026amp; Information Scientist — Hakoona Jul 2023 – Present · Amsterdam (Hybrid)\nLeading the technical vision for Hakoona\u0026rsquo;s semantic platforms. Running experiments that translate AI and (hyper)graph technology into production-ready software components. Building linguistic models, semantic engines, and LLM applications. Member of the DataSingularity R\u0026amp;D ThinkTank since 2020.\nAI \u0026amp; Data Engineer — Ookull 2022 – Present · Netherlands (Remote)\nInnovation, AI, and knowledge graph solutions. Python-based systems leveraging graph technologies and large language models.\nData Science Educator — Fenyx IT Academy Sep 2018 – Present · Amsterdam (Volunteer)\nHelping talented newcomer refugees gain Python and Data Science skills and launch IT careers in the Netherlands. Curriculum: Python, data engineering, ML, visualization, real-world projects.\nEarlier: 17 years as Microsoft Certified Trainer (2008–2024) · Lecturer in Cyber Threats and Defense at Gazi University (2014–2016) · Remote Sensing Division Manager at the Turkish State Meteorological Service (2005–2014), responsible for the national radar/satellite data network as a member of EUMETSAT.\nRecent clients # Client Sector Engagement Hakoona AI / Knowledge Graphs AI \u0026amp; data engineering — ongoing Ookull AI / Innovation AI \u0026amp; data engineering — ongoing Vynova Group Chemicals / Manufacturing Data science \u0026amp; training PWN Water Management AI project Wetterskip Fryslân Water Authority Training project UAF Non-profit / Education Training project How we work # Pragmatic engineering — choosing the right tool for the job, not the most fashionable one. Clean, documented, testable code that your team can maintain long after the engagement ends.\nFlexibility — short sprints or long-term embedded roles Autonomy — hit the ground running, minimal hand-holding Transparency — clear communication, regular updates Quality — we don\u0026rsquo;t ship code we wouldn\u0026rsquo;t want to maintain ourselves Certifications # Microsoft Certified Trainer (MCT) — 2008–2024. 17 years delivering official Microsoft technical training.\nActive Microsoft Azure \u0026amp; AI:\nFabric Data Engineer Associate · valid Feb 2027 Azure AI Engineer Associate · valid Jan 2027 Fabric Analytics Engineer Associate · valid May 2027 Azure Data Scientist Associate · valid Feb 2027 Azure Administrator Associate · valid Sep 2027 Graph Databases: Neo4j Certified Professional (2025) — GraphRAG, hypergraph structures, Cypher, graph data modelling, Neo4j in production.\nData Science: IBM Data Science Professional Certificate (2023) · Certified Associate in Python Programming, PCAP (2019).\nLanguages # 🇳🇱 Dutch — Professional working proficiency · 🇬🇧 English — Fluent · 🇹🇷 Turkish — Native\nLet\u0026rsquo;s work together # Interested in working together? Send a message or connect on LinkedIn.\n","externalUrl":null,"permalink":"/about/","section":"Demir Technology","summary":"","title":"About","type":"page"},{"content":"","externalUrl":null,"permalink":"/authors/","section":"Authors","summary":"","title":"Authors","type":"authors"},{"content":"Demir Technology is currently available for new projects — both short-term engagements and longer embedded roles.\nThe best way to reach us is by email. We typically respond within one business day.\nGet In Touch # Email: admin@demirtech.eu\nLinkedIn: linkedin.com/in/fatih-d-91b861137\nLocation # Based in the Netherlands 🇳🇱\nAvailable for:\nRemote work (anywhere in Europe) On-site (Netherlands, and other EU locations by arrangement) What to Include in Your Message # To get a useful response quickly, please share:\nA brief description of the project or challenge The timeline you have in mind Whether you\u0026rsquo;re looking for project-based or extended freelance work Your preferred stack or technology constraints (if any) Availability # Current status Available Earliest start Immediately Preferred engagement 3+ months Time zone CET (UTC+1/+2) Demir Technology is registered at the Dutch Chamber of Commerce (KvK).\n","externalUrl":null,"permalink":"/contact/","section":"Demir Technology","summary":"","title":"Contact","type":"page"},{"content":" Privacy Policy # This website (demirtech.eu) is operated by Demir Technology, registered in the Netherlands (KvK).\nData collected: This website does not use cookies or collect personal data beyond what you voluntarily submit via email contact.\nContact form / email: Any information you send to info@demirtech.eu is used solely to respond to your inquiry and is not shared with third parties.\nHosting: This site is hosted on site.eu infrastructure located within the EU.\nFor questions about this policy, contact: admin@demirtech.eu\nLast updated: May 2025\n","externalUrl":null,"permalink":"/privacy/","section":"Demir Technology","summary":"","title":"Privacy Policy","type":"page"},{"content":"A selection of recent freelance engagements. Client names are kept confidential where requested. Each case study includes a high-level architecture sketch and the measured outcomes.\nReal-Time E-Commerce Analytics Platform # Industry: E-Commerce · Engagement: ~3 months · Stack: Kafka · Spark Structured Streaming · Databricks · Power BI\nRedesigned a batch-based reporting system into a real-time analytics platform for a mid-sized Dutch e-commerce company. The new platform processes over 2 million events per day with sub-minute latency.\nflowchart LR src1[Web App]:::src --\u003e kafka src2[Orders DB]:::src --\u003e kafka src3[Inventory]:::src --\u003e kafka src4[Payments]:::src --\u003e kafka kafka[(Apache Kafka)]:::core --\u003e spark[SparkStructured Streaming]:::core spark --\u003e dbx[(DatabricksDelta Lake)]:::store dbx --\u003e pbi[Power BIDashboards]:::out pbi --\u003e users([Business Team]):::user classDef src fill:#1e293b,stroke:#475569,color:#e2e8f0 classDef core fill:#1e40af,stroke:#3b82f6,color:#fff classDef store fill:#155e75,stroke:#06b6d4,color:#fff classDef out fill:#0f766e,stroke:#14b8a6,color:#fff classDef user fill:#3b0764,stroke:#a855f7,color:#fff Key outcomes:\nReduced reporting lag from 24 hours to under 60 seconds Unified 5 disconnected data sources into a single streaming pipeline Enabled real-time inventory and sales dashboards for the business team LLM-Powered Document Intelligence System # Industry: Legal / Professional Services · Engagement: ~4 months · Stack: Python · LangChain · OpenAI · PostgreSQL (pgvector) · FastAPI\nBuilt a Retrieval-Augmented Generation (RAG) system that lets legal professionals query large document repositories in natural language. Integrated with the firm\u0026rsquo;s existing document management workflows.\nflowchart LR docs[LegalDocuments]:::src --\u003e emb[EmbeddingPipeline]:::core emb --\u003e pg[(PostgreSQL+ pgvector)]:::store user([Lawyer]):::user --\u003e api[FastAPI]:::core api --\u003e lc[LangChainRAG Engine]:::core lc \u003c--\u003e pg lc \u003c--\u003e llm[OpenAIGPT-4]:::out api --\u003e user classDef src fill:#1e293b,stroke:#475569,color:#e2e8f0 classDef core fill:#1e40af,stroke:#3b82f6,color:#fff classDef store fill:#155e75,stroke:#06b6d4,color:#fff classDef out fill:#0f766e,stroke:#14b8a6,color:#fff classDef user fill:#3b0764,stroke:#a855f7,color:#fff Key outcomes:\nReduced document review time by ~60% for participating teams Achieved 92% answer accuracy on internal benchmark dataset Deployed on Azure with a scalable, cost-efficient architecture Cloud Data Warehouse Migration # Industry: Logistics · Engagement: ~6 months · Stack: Python · dbt · Airflow · Snowflake · Terraform\nLed the migration of a legacy on-premise Oracle data warehouse to Snowflake for a logistics company. Rebuilt 80+ reports and redesigned the dimensional model from scratch using dbt.\nflowchart LR ora[(Oracleon-prem)]:::legacy --\u003e mig[MigrationPipeline]:::core mig --\u003e snow[(Snowflake)]:::store snow --\u003e dbt[dbtTransformations]:::core dbt --\u003e mart[Star SchemaMarts]:::store mart --\u003e rpt[80+ Reports]:::out air[Airflow]:::orch -.orchestrates.-\u003e mig air -.orchestrates.-\u003e dbt tf[Terraform IaC]:::orch -.provisions.-\u003e snow classDef legacy fill:#3f1d1d,stroke:#dc2626,color:#fecaca classDef core fill:#1e40af,stroke:#3b82f6,color:#fff classDef store fill:#155e75,stroke:#06b6d4,color:#fff classDef out fill:#0f766e,stroke:#14b8a6,color:#fff classDef orch fill:#3b0764,stroke:#a855f7,color:#fff Key outcomes:\nCut query times by an average of 78% Reduced infrastructure cost by 40% year-over-year Onboarded analytics team to dbt — full test coverage in 6 weeks Churn Prediction ML Pipeline # Industry: SaaS / Fintech · Engagement: ~3 months · Stack: Python · scikit-learn · MLflow · Airflow · AWS SageMaker\nDeveloped an end-to-end churn prediction system for a B2B SaaS company — including automated retraining, drift detection, and integration with the CRM for proactive customer outreach.\nflowchart LR src[(Product +Billing DB)]:::store --\u003e feat[FeatureEngineering]:::core feat --\u003e train[scikit-learnTraining]:::core train --\u003e mlf[(MLflowRegistry)]:::store mlf --\u003e sage[SageMakerEndpoint]:::core sage --\u003e sf[Salesforce CRM]:::out sf --\u003e csm([Customer Success]):::user air[Airflow]:::orch -.weekly retrain.-\u003e train air -.drift detection.-\u003e sage classDef src fill:#1e293b,stroke:#475569,color:#e2e8f0 classDef core fill:#1e40af,stroke:#3b82f6,color:#fff classDef store fill:#155e75,stroke:#06b6d4,color:#fff classDef out fill:#0f766e,stroke:#14b8a6,color:#fff classDef user fill:#3b0764,stroke:#a855f7,color:#fff classDef orch fill:#7c2d12,stroke:#f97316,color:#fff Key outcomes:\nModel AUC-ROC: 0.89 on holdout set Automated weekly retraining pipeline with data quality gates Predictions integrated directly into Salesforce via API Modern Data Stack Implementation (Startup) # Industry: Health Tech · Engagement: 6 weeks · Stack: Airbyte · dbt · BigQuery · Metabase · GitHub Actions\nSet up a complete data platform from zero for a health tech startup — from raw ingestion to business dashboards — in six weeks.\nflowchart LR s1[Stripe]:::src --\u003e ab s2[Postgres]:::src --\u003e ab s3[Salesforce]:::src --\u003e ab s4[12+ othersources]:::src --\u003e ab ab[AirbyteConnectors]:::core --\u003e bq[(BigQueryWarehouse)]:::store bq --\u003e dbt[dbt models+ tests]:::core dbt --\u003e meta[MetabaseDashboards]:::out meta --\u003e teams([Product +Leadership]):::user gh[GitHub Actions]:::orch -.CI/CD.-\u003e dbt classDef src fill:#1e293b,stroke:#475569,color:#e2e8f0 classDef core fill:#1e40af,stroke:#3b82f6,color:#fff classDef store fill:#155e75,stroke:#06b6d4,color:#fff classDef out fill:#0f766e,stroke:#14b8a6,color:#fff classDef user fill:#3b0764,stroke:#a855f7,color:#fff classDef orch fill:#7c2d12,stroke:#f97316,color:#fff Key outcomes:\n15+ data sources connected via Airbyte Fully documented dbt project with 100% test coverage on critical models Self-serve Metabase dashboards adopted by all product and leadership teams Looking for similar results? # I\u0026rsquo;d love to discuss how Demir Technology can help with your data or AI challenge. Get in touch or view the services for engagement details.\n","externalUrl":null,"permalink":"/projects/","section":"Demir Technology","summary":"","title":"Projects","type":"page"},{"content":"","externalUrl":null,"permalink":"/series/","section":"Series","summary":"","title":"Series","type":"series"},{"content":" Demir Technology offers a focused set of services covering the full data journey — from raw ingestion to ontology-instructed AI systems. Engagements span Microsoft Fabric data platforms, knowledge-graph-powered AI, and independent advisory backed by 25+ years of hands-on experience. Data Engineering \u0026amp; Microsoft Fabric Production-grade data infrastructure your team can trust and maintain. We architect unified data estates on Microsoft Fabric and design real-time pipelines that survive contact with production. What we deliver ETL/ELT pipelines that scale with your data volume Real-time streaming with Kafka and Flink Orchestration with Airflow, Prefect, or Dagster Data quality frameworks (Great Expectations, dbt tests) Lakehouse architecture on Delta Lake or Apache Iceberg DataOps: version control, CI/CD for data, automated testing Typical stack Microsoft Fabric dbt Apache Spark Kafka Flink Airflow Snowflake Databricks Terraform AI \u0026amp; Semantic Systems Beyond standard models — we design ontology-instructed AI: knowledge graphs integrated with LLMs to create semantic engines that understand business logic, not just process text. From prototype to production, reliably. What we deliver Knowledge graph design and GraphRAG with Neo4j LLM \u0026amp; RAG applications (OpenAI, Hugging Face, LangChain) End-to-end ML pipelines: features, training, evaluation, serving MLOps: experiment tracking, model registry, CI/CD for ML Recommendation systems, forecasting, NLP, and computer vision AI strategy: identifying the highest-value, lowest-risk use cases Typical stack Python PyTorch scikit-learn LangChain OpenAI API Hugging Face Neo4j Cypher MLflow Vertex AI Consulting, Architecture \u0026amp; Training Not sure where to start — or about to make a major investment? Independent advisory, architecture reviews, and MCT-led training to cut through the noise before you commit. What we deliver Data platform audits: assess your setup, identify bottlenecks Technology selection: unbiased evaluation that fits your budget Architecture reviews: a second opinion before a big investment Hands-on workshops on Microsoft Fabric, Azure AI, and the modern data stack Team upskilling: dbt, Spark, Airflow, MLOps, Neo4j Strategic AI advisory: portfolio prioritisation and roadmaps Credentials MCT (2008–2024) Azure AI Engineer Fabric Data Engineer Fabric Analytics Engineer Azure Data Scientist Neo4j Certified Also available # We also support Data Analytics \u0026amp; Business Intelligence workstreams when they sit alongside a platform engagement — semantic layer design, BI dashboards (Power BI, Looker, Metabase, Grafana), KPI frameworks, and self-serve analytics setup so analysts move fast without engineering bottlenecks.\nEngagement models # Model Best for Project-based A specific, scoped deliverable with a clear start and end Extended freelance Embedded in your team for weeks or months Part-time retainer Ongoing support for a fixed number of days per week Consulting day One-day advisory session or architecture review Workshop / Training 1–3 day MCT-led training on Fabric, Azure AI, or the modern data stack Both remote and on-site (Netherlands and Europe).\nLet\u0026rsquo;s talk # Ready to get started, or not sure which service fits your situation? Get in touch — the first conversation is always free.\n","externalUrl":null,"permalink":"/services/","section":"Demir Technology","summary":"","title":"Services","type":"page"}]