Halide Yılmaz: METU Computer Engineering graduate and one of the co-founders of HatsonTech and Caseon AI, writing about what she learns on machine learning, data and AI governance.https://halideyilmaz.com/enhttps://halideyilmaz.com/en/posts/what-is-machine-learning/https://halideyilmaz.com/en/posts/what-is-machine-learning/A plain-language guide to machine learning: learning from examples, supervised and unsupervised learning, training versus prediction, and what ML is not.Fri, 06 Nov 2026 06:17:00 GMTMachine LearningArtificial IntelligenceFundamentalshttps://halideyilmaz.com/en/posts/eu-ai-act-timeline-2026/https://halideyilmaz.com/en/posts/eu-ai-act-timeline-2026/The EU AI Act was not delayed; part of it was. What Regulation (EU) 2026/1744 actually changed, why Article 50 transparency has applied since 2 August 2026, and how the 2 December 2027 and 2 August 2028 deadlines now work, traced to primary sources.Thu, 13 Aug 2026 20:29:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/high-risk-ai-system-requirements/https://halideyilmaz.com/en/posts/high-risk-ai-system-requirements/The Article 9-15 obligations for high-risk AI systems, translated from legal language into engineering deliverables: a risk register, data governance, Annex IV technical documentation, event logging, an accuracy declaration and a human oversight design.Thu, 13 Aug 2026 20:28:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/ai-content-labeling-transparency/https://halideyilmaz.com/en/posts/ai-content-labeling-transparency/Article 50 transparency obligations took effect on 2 August 2026 and were not deferred. What machine-readable marking actually means, hard binding versus soft binding, C2PA content credentials, and why a watermark usually dies at the first re-encode.Thu, 13 Aug 2026 20:27:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/iso-42001-ai-management-system/https://halideyilmaz.com/en/posts/iso-42001-ai-management-system/What ISO/IEC 42001 actually requires, how its Annex A controls map onto real ML artefacts, what ISO/IEC 42005 and 42006 added, and why the standard is not harmonised and therefore grants no presumption of conformity under the EU AI Act.Thu, 13 Aug 2026 20:26:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/nist-ai-rmf-in-practice/https://halideyilmaz.com/en/posts/nist-ai-rmf-in-practice/Instead of restating the four functions of the NIST AI RMF, this post turns them into engineering work. Why the MEASURE function is the one most teams leave empty, what the Generative AI Profile added, and why the framework is currently under revision.Thu, 13 Aug 2026 20:25:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/ai-impact-assessment-fria/https://halideyilmaz.com/en/posts/ai-impact-assessment-fria/A fundamental rights impact assessment, an ISO/IEC 42005 impact assessment and a data protection impact assessment are three different documents. Who each one binds, when it is required, when it applies, and how to run them together without doing the work three times.Thu, 13 Aug 2026 20:24:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/fairness-auditing-in-practice/https://halideyilmaz.com/en/posts/fairness-auditing-in-practice/The operational version of a fairness audit: what to do when you may not hold the protected attribute, how to choose a metric knowing the impossibility results, how large a subgroup must be before the estimate means anything, and how to write the report.Thu, 13 Aug 2026 20:23:00 GMTEthicsAIhttps://halideyilmaz.com/en/posts/ai-incident-reporting-post-market/https://halideyilmaz.com/en/posts/ai-incident-reporting-post-market/Serious incident reporting runs on a 15-day clock, 10 days in the event of death and 2 days for a widespread infringement. What logging infrastructure, severity taxonomy and on-call path an ML team needs in order to meet those clocks at all.Thu, 13 Aug 2026 20:22:00 GMTProductionAIhttps://halideyilmaz.com/en/posts/data-provenance-training-data-licensing/https://halideyilmaz.com/en/posts/data-provenance-training-data-licensing/How do you record training-data provenance through a preprocessing pipeline? The sharp rise in robots.txt restrictions, the GPAI training-data summary template, where the IETF work stands, and what the 2025-2026 copyright rulings mean for engineering practice.Thu, 13 Aug 2026 20:21:00 GMTDataAIhttps://halideyilmaz.com/en/posts/data-contracts-for-ml/https://halideyilmaz.com/en/posts/data-contracts-for-ml/A silent upstream change degrades a model without tripping a single data alert. What a data contract is, which sections ODCS defines, how to enforce one in CI, and why the tooling comparison every ranking page repeats is now out of date on ownership and licensing.Thu, 13 Aug 2026 20:20:00 GMTDataAIhttps://halideyilmaz.com/en/posts/datasheets-for-datasets/https://halideyilmaz.com/en/posts/datasheets-for-datasets/A model card documents the model, not the dataset. Which questions a datasheet answers, how it maps onto Annex IV technical documentation, and why this is the one document that cannot be produced retroactively.Thu, 13 Aug 2026 20:19:00 GMTDataAIhttps://halideyilmaz.com/en/posts/annotator-agreement-evaluator-reliability/https://halideyilmaz.com/en/posts/annotator-agreement-evaluator-reliability/Raw percentage agreement is not reliability. How Cohen kappa, Fleiss kappa and Krippendorff alpha correct for chance, how to read the thresholds, and how to validate an LLM judge the way you validate a human annotator.Thu, 13 Aug 2026 20:18:00 GMTMetricsAIhttps://halideyilmaz.com/en/posts/benchmark-contamination/https://halideyilmaz.com/en/posts/benchmark-contamination/Test set leakage is no longer one team's cross-validation mistake but a defect in the whole field's evaluation infrastructure. How contamination and saturation inflate scores, and how to build your own private held-out set.Thu, 13 Aug 2026 20:17:00 GMTEvaluationAIhttps://halideyilmaz.com/en/posts/statistical-significance-model-comparison/https://halideyilmaz.com/en/posts/statistical-significance-model-comparison/One model beating another by a point is usually noise. How test set sampling error, bootstrap confidence intervals, paired tests and multiple comparison correction tell you whether the gap between two models is real.Thu, 13 Aug 2026 20:16:00 GMTEvaluationAIhttps://halideyilmaz.com/en/posts/model-calibration-probabilities/https://halideyilmaz.com/en/posts/model-calibration-probabilities/An uncalibrated score makes threshold selection meaningless and silently invalidates every downstream cost-benefit decision. Reliability diagrams, expected calibration error, Platt scaling, isotonic and temperature scaling explained.Thu, 13 Aug 2026 20:15:00 GMTMetricsAIhttps://halideyilmaz.com/en/posts/conformal-prediction-uncertainty/https://halideyilmaz.com/en/posts/conformal-prediction-uncertainty/Conformal prediction turns any model into prediction sets with a distribution-free, finite-sample coverage guarantee. Split conformal step by step, the nonconformity score, and why marginal coverage is not conditional coverage.Thu, 13 Aug 2026 20:14:00 GMTMethodsAIhttps://halideyilmaz.com/en/posts/is-your-model-anonymous-memorization/https://halideyilmaz.com/en/posts/is-your-model-anonymous-memorization/Does your model still hold personal data? A protocol that maps the EDPB three-criteria test onto measurable attack surfaces: membership inference, training-data extraction and memorization capacity, with what to report.Thu, 13 Aug 2026 20:13:00 GMTPrivacyAIhttps://halideyilmaz.com/en/posts/embedding-drift-monitoring/https://halideyilmaz.com/en/posts/embedding-drift-monitoring/Why a per-dimension KS test fails on a 768-dimensional embedding, what to measure instead, how an embedding model version bump silently invalidates your reference window, and why drift is not degradation.Thu, 13 Aug 2026 20:12:00 GMTProductionAIhttps://halideyilmaz.com/en/posts/performance-estimation-without-labels/https://halideyilmaz.com/en/posts/performance-estimation-without-labels/If the labels arrive in ninety days, is your model broken now? How confidence-based performance estimation works, the assumptions it rests on, and why it goes quiet in exactly the scenario people reach for it.Thu, 13 Aug 2026 20:11:00 GMTProductionAIhttps://halideyilmaz.com/en/posts/feature-stores-training-serving-skew/https://halideyilmaz.com/en/posts/feature-stores-training-serving-skew/Offline the model scores well, in production it behaves differently. The four concrete sources of training-serving skew, why point-in-time correctness is really leakage prevention, and when a feature store earns its operational cost.Thu, 13 Aug 2026 20:10:00 GMTProductionAIhttps://halideyilmaz.com/en/posts/causal-inference-uplift-modeling/https://halideyilmaz.com/en/posts/causal-inference-uplift-modeling/A churn model that perfectly predicts who will leave tells you nothing about whom to call. Uplift modelling, the four segments, confounders versus colliders, what to do when you cannot randomise, and why reading SHAP causally is a mistake.Thu, 13 Aug 2026 20:09:00 GMTMethodsAIhttps://halideyilmaz.com/en/posts/machine-unlearning-right-to-be-forgotten/https://halideyilmaz.com/en/posts/machine-unlearning-right-to-be-forgotten/Making a model forget a person is nothing like deleting a row from a database. Exact versus approximate unlearning, relearning attacks that bring forgotten behaviour back, the proof problem, and the data architecture that actually works.Thu, 13 Aug 2026 20:08:00 GMTPrivacyAIhttps://halideyilmaz.com/en/posts/time-series-foundation-models/https://halideyilmaz.com/en/posts/time-series-foundation-models/The answer flipped and most published content has not caught up. What fev-bench and GIFT-Eval actually report, why LLM-adapted forecasters are a separate story, and how to run the comparison honestly on your own series.Thu, 13 Aug 2026 20:07:00 GMTMethodsAIhttps://halideyilmaz.com/en/posts/turkish-nlp-data-quality-benchmarks/https://halideyilmaz.com/en/posts/turkish-nlp-data-quality-benchmarks/The Turkish resource landscape exists mostly as scattered GitHub lists and arXiv papers. What actually exists, what each benchmark measures, what the Cetvel finding says about data quality, and how to build your own evaluation set.Thu, 13 Aug 2026 20:06:00 GMTTurkish NLPAIhttps://halideyilmaz.com/en/posts/data-centric-ai/https://halideyilmaz.com/en/posts/data-centric-ai/Data-centric AI holds the model fixed and improves the dataset instead. How label consistency, coverage gaps and error analysis pull more accuracy out of the same architecture, explained with practical examples.Sun, 09 Aug 2026 20:59:00 GMTDataAIhttps://halideyilmaz.com/en/posts/class-imbalance/https://halideyilmaz.com/en/posts/class-imbalance/In fraud, failure or rare-disease problems the positive class is tiny. How resampling, class weights, threshold tuning and the precision-recall curve let you build a model that actually works on imbalanced data.Sun, 09 Aug 2026 20:58:00 GMTEvaluationAIhttps://halideyilmaz.com/en/posts/active-learning/https://halideyilmaz.com/en/posts/active-learning/Labelling everything is expensive. Active learning picks the examples a model is least sure about, buying more accuracy from the same budget. Uncertainty sampling, diversity, and the cold-start problem.Sun, 09 Aug 2026 20:57:00 GMTDataAIhttps://halideyilmaz.com/en/posts/data-leakage-cross-validation/https://halideyilmaz.com/en/posts/data-leakage-cross-validation/A validation score that looks too good is usually a sign of leakage. Temporal leakage, group leakage, scaling mistakes, and how to choose the right cross-validation scheme.Sun, 09 Aug 2026 20:56:00 GMTEvaluationAIhttps://halideyilmaz.com/en/posts/bias-and-fairness/https://halideyilmaz.com/en/posts/bias-and-fairness/Bias usually comes from the data and the problem definition rather than the model. Historical bias, representation gaps, fairness metrics, and why those metrics contradict each other.Sun, 09 Aug 2026 20:55:00 GMTEthicsAIhttps://halideyilmaz.com/en/posts/anomaly-detection/https://halideyilmaz.com/en/posts/anomaly-detection/Anomaly detection usually has no labels and very few positives. Statistical methods, Isolation Forest, autoencoders, and how to manage the false-alarm rate through threshold choice.Sun, 09 Aug 2026 20:54:00 GMTMethodsAIhttps://halideyilmaz.com/en/posts/time-series-forecasting/https://halideyilmaz.com/en/posts/time-series-forecasting/Time series problems obey different rules from ordinary machine learning. Trend, seasonality, stationarity, baselines and walk-forward validation, explained practically.Sun, 09 Aug 2026 20:53:00 GMTMethodsAIhttps://halideyilmaz.com/en/posts/recommender-systems/https://halideyilmaz.com/en/posts/recommender-systems/Recommenders rest on two ideas: similar users and similar items. Collaborative filtering, content-based methods, hybrids, the cold-start problem and the diversity trade-off.Sun, 09 Aug 2026 20:52:00 GMTMethodsAIhttps://halideyilmaz.com/en/posts/privacy-preserving-learning/https://halideyilmaz.com/en/posts/privacy-preserving-learning/You can train a model without centralising the data. Federated learning, differential privacy, the privacy budget, and what these methods actually guarantee.Sun, 09 Aug 2026 20:51:00 GMTPrivacyAIhttps://halideyilmaz.com/en/posts/model-cards-governance/https://halideyilmaz.com/en/posts/model-cards-governance/A model card records what a model does, where it fails, and what data trained it. A practical look at governance, auditability and risk classification.Sun, 09 Aug 2026 20:50:00 GMTGovernanceAIhttps://halideyilmaz.com/en/posts/cnn-image-processing-basics/https://halideyilmaz.com/en/posts/cnn-image-processing-basics/An intuitive guide to CNNs through filters, convolution, pooling and layers: how a machine actually parses and "sees" an image, with a short code sketch.Sun, 09 Aug 2026 20:49:00 GMTVisionAIhttps://halideyilmaz.com/en/posts/rnn-lstm-to-transformer/https://halideyilmaz.com/en/posts/rnn-lstm-to-transformer/An intuitive guide to the evolution of sequence architectures: how RNNs work, their limits like vanishing gradients and the inability to parallelize, the gates of the LSTM, and why attention won out with the Transformer.Sun, 09 Aug 2026 20:48:00 GMTHistoryAIhttps://halideyilmaz.com/en/posts/overfitting-regularization/https://halideyilmaz.com/en/posts/overfitting-regularization/Why do models memorize? An intuitive guide to overfitting and regularization: dropout, L1/L2, early stopping, and cross-validation explained with everyday examples.Sun, 09 Aug 2026 20:47:00 GMTFundamentalsAIhttps://halideyilmaz.com/en/posts/gradient-descent-optimization/https://halideyilmaz.com/en/posts/gradient-descent-optimization/An intuitive walk through gradient descent via the loss function, learning rate, SGD and Adam: how does a model really "learn" by measuring and reducing its error?Sun, 09 Aug 2026 20:46:00 GMTTrainingAIhttps://halideyilmaz.com/en/posts/classification-metrics/https://halideyilmaz.com/en/posts/classification-metrics/Starting from the confusion matrix, we explain accuracy, precision, recall, F1, and ROC-AUC through everyday examples and clarify which metric is the right choice and when.Sun, 09 Aug 2026 20:45:00 GMTMetricsAIhttps://halideyilmaz.com/en/posts/unsupervised-learning-clustering/https://halideyilmaz.com/en/posts/unsupervised-learning-clustering/An intuitive guide to unsupervised learning through k-means, hierarchical clustering and PCA: how to extract hidden patterns from unlabeled data and choose the number of clusters.Sun, 09 Aug 2026 20:44:00 GMTLearningAIhttps://halideyilmaz.com/en/posts/feature-engineering/https://halideyilmaz.com/en/posts/feature-engineering/An intuitive guide to feature engineering: deriving meaningful features from raw data, scaling, encoding, and the data-leakage traps that make models collapse in production.Sun, 09 Aug 2026 20:43:00 GMTDataAIhttps://halideyilmaz.com/en/posts/model-interpretability-xai/https://halideyilmaz.com/en/posts/model-interpretability-xai/An intuitive guide to SHAP, LIME, and attention visualization: why a model decides the way it does, and how these explanations strengthen trust and accountability.Sun, 09 Aug 2026 20:42:00 GMTInterpretabilityAIhttps://halideyilmaz.com/en/posts/synthetic-data-generation/https://halideyilmaz.com/en/posts/synthetic-data-generation/Synthetic data is generated data that imitates the patterns of reality. We explain, with intuitive examples, how it overcomes data scarcity, protects privacy, how LLMs generate it, and the risk of model collapse.Sun, 09 Aug 2026 20:41:00 GMTDataAIhttps://halideyilmaz.com/en/posts/model-monitoring-drift/https://halideyilmaz.com/en/posts/model-monitoring-drift/A practical guide to data and concept drift in production ML models: detecting performance decay, retraining triggers, and keeping your production ML healthy.Sun, 09 Aug 2026 20:40:00 GMTProductionAIhttps://halideyilmaz.com/en/posts/attention-mechanism-explained/https://halideyilmaz.com/en/posts/attention-mechanism-explained/When a model understands a word, which words does it "look at"? An intuitive guide to Query-Key-Value, self-attention, and the Transformer revolution.Sun, 09 Aug 2026 20:39:00 GMTDeepAIhttps://halideyilmaz.com/en/posts/tokenization-turkish-challenges/https://halideyilmaz.com/en/posts/tokenization-turkish-challenges/Why does Turkish consume more tokens? Agglutination, vocabulary explosion, BPE and SentencePiece, and how efficient tokenization affects cost and speed.Sun, 09 Aug 2026 20:38:00 GMTTurkish NLPAIhttps://halideyilmaz.com/en/posts/embedding-models-semantic-similarity/https://halideyilmaz.com/en/posts/embedding-models-semantic-similarity/An intuitive guide to turning text into vectors, cosine similarity, choosing an embedding model, and multilingual embeddings, with everyday analogies and a short code example.Sun, 09 Aug 2026 20:37:00 GMTFundamentalsAIhttps://halideyilmaz.com/en/posts/rag-chunking-strategies/https://halideyilmaz.com/en/posts/rag-chunking-strategies/RAG chunking strategies: fixed size, sentence/paragraph, overlap, semantic chunking, and preserving heading context. How splitting affects quality, explained intuitively.Sun, 09 Aug 2026 20:36:00 GMTRAGAIhttps://halideyilmaz.com/en/posts/reranking-hybrid-search/https://halideyilmaz.com/en/posts/reranking-hybrid-search/An intuitive guide to hybrid search blending BM25 and semantic search, score fusion with RRF, and reranking with a cross-encoder, with short code examples.Sun, 09 Aug 2026 20:35:00 GMTRetrievalAIhttps://halideyilmaz.com/en/posts/context-window-long-documents/https://halideyilmaz.com/en/posts/context-window-long-documents/What is a context window? Token limits, the "lost in the middle" problem, long context vs RAG, and summarization chains (map-reduce, refine) explained with intuitive analogies.Sun, 09 Aug 2026 20:34:00 GMTLimitsAIhttps://halideyilmaz.com/en/posts/llm-agents-tool-use/https://halideyilmaz.com/en/posts/llm-agents-tool-use/How do LLM agents call tools, plan, and run multi-step tasks? We explain MCP, the opportunities, and the risks with plain, everyday analogies.Sun, 09 Aug 2026 20:33:00 GMTAgentsAIhttps://halideyilmaz.com/en/posts/knowledge-graph-rag/https://halideyilmaz.com/en/posts/knowledge-graph-rag/GraphRAG: retrieval that combines a knowledge graph with RAG. Connections between entities, multi-hop and global questions, and closing classic RAG's gaps, explained intuitively.Sun, 09 Aug 2026 20:32:00 GMTAdvanced RAGAIhttps://halideyilmaz.com/en/posts/model-inference-scaling/https://halideyilmaz.com/en/posts/model-inference-scaling/A practical guide to latency, throughput, quantization, KV-cache and batching, with everyday analogies, focused on lowering the cost of GPU inference.Sun, 09 Aug 2026 20:31:00 GMTScalingAIhttps://halideyilmaz.com/en/posts/data-quality-labeling/https://halideyilmaz.com/en/posts/data-quality-labeling/Garbage in, garbage out: data cleaning, labeling strategies, synthetic data, and honest evaluation sets that make up a model's real fuel.Sun, 09 Aug 2026 20:30:00 GMTDataAIhttps://halideyilmaz.com/en/posts/who-is-halide-yilmaz/https://halideyilmaz.com/en/posts/who-is-halide-yilmaz/Halide Yılmaz is a METU Computer Engineering graduate and one of the co-founders of HatsonTech and Caseon AI, building companies in AI and software.Sun, 09 Aug 2026 20:05:00 GMTProfileEntrepreneurshipHatsonTechCaseon AIhttps://halideyilmaz.com/en/posts/how-hatsontech-started/https://halideyilmaz.com/en/posts/how-hatsontech-started/HatsonTech is a software and AI company founded in Ankara in August 2026. A short account of how it started, what it does, and how it relates to Caseon AI, our legal-technology company.Sun, 09 Aug 2026 20:04:00 GMTCompanyHatsonTechCaseon AIEntrepreneurship