İsmail Tarık Şenkal: developer working on Turkish-focused artificial intelligence, large language models, RAG and natural language processing. Co-founder of HatsonTech.https://tariksenkal.com/enhttps://tariksenkal.com/en/posts/what-is-cop31-antalya/https://tariksenkal.com/en/posts/what-is-cop31-antalya/COP31 runs 9–20 November 2026 in Antalya. What a COP is, how the Türkiye–Australia presidency split works, what is on the table and why it matters.Fri, 06 Nov 2026 06:17:00 GMTClimateCOP31Türkiyehttps://tariksenkal.com/en/posts/prefix-caching-economics/https://tariksenkal.com/en/posts/prefix-caching-economics/Cache reads cost 0.1x, writes 1.25x. Break-even is two requests. So why is your cache silently doing nothing? Minimum prefixes, the 20-block rule and the parallel trap.Thu, 13 Aug 2026 20:59:00 GMTCostAIhttps://tariksenkal.com/en/posts/speculative-decoding/https://tariksenkal.com/en/posts/speculative-decoding/Draft models, EAGLE, MTP, n-gram and suffix decoding. Why acceptance rate is the only metric that matters, how to configure vLLM, and when speculation does not help.Thu, 13 Aug 2026 20:58:00 GMTInfrastructureAIhttps://tariksenkal.com/en/posts/quantization-in-production/https://tariksenkal.com/en/posts/quantization-in-production/How to choose between FP8, NVFP4, AWQ and GPTQ: the hardware support matrix, serving with vLLM, and why low-bit quantization degrades non-English output first.Thu, 13 Aug 2026 20:57:00 GMTEfficiencyAIhttps://tariksenkal.com/en/posts/context-compaction-and-reset/https://tariksenkal.com/en/posts/context-compaction-and-reset/What to do when the context window fills up: context editing, the clear_tool_uses strategy, server-side compaction and full context resets — with a decision rule.Thu, 13 Aug 2026 20:56:00 GMTArchitectureAIhttps://tariksenkal.com/en/posts/prompt-injection-defense-by-design/https://tariksenkal.com/en/posts/prompt-injection-defense-by-design/Better prompts do not solve prompt injection. CaMeL, the dual-LLM pattern, capability-based policy and the lethal trifecta — defences enforced outside the model.Thu, 13 Aug 2026 20:55:00 GMTSecurityAIhttps://tariksenkal.com/en/posts/mcp-apps-generative-ui/https://tariksenkal.com/en/posts/mcp-apps-generative-ui/The MCP Apps extension turns tool results into interactive UI. ui:// resources, the _meta.ui field, sandboxed iframes and the postMessage protocol, with real code.Thu, 13 Aug 2026 20:54:00 GMTProductAIhttps://tariksenkal.com/en/posts/a2a-agent-interoperability/https://tariksenkal.com/en/posts/a2a-agent-interoperability/A2A v1.0 standardises agent discovery and task delegation. Agent Cards, the task lifecycle, how it differs from MCP, and the agentic commerce layer above it.Thu, 13 Aug 2026 20:53:00 GMTToolsAIhttps://tariksenkal.com/en/posts/multi-agent-cost-reality/https://tariksenkal.com/en/posts/multi-agent-cost-reality/When multi-agent architectures win and when they burn money: reconciling two primary sources, the token multiplier, failure modes and a concrete decision table.Thu, 13 Aug 2026 20:52:00 GMTAgentsAIhttps://tariksenkal.com/en/posts/data-erasure-rag-and-weights/https://tariksenkal.com/en/posts/data-erasure-rag-and-weights/A deletion request touches six places: source, chunks, vectors, caches, logs and fine-tuned weights. A practical erasure strategy and schema for each of them.Thu, 13 Aug 2026 20:51:00 GMTApplicationsAIhttps://tariksenkal.com/en/posts/does-llms-txt-work/https://tariksenkal.com/en/posts/does-llms-txt-work/A 137,210-domain server-log study says llms.txt is barely fetched and Google says it does not use it. So what did v2 change, and what actually moves AI citations?Thu, 13 Aug 2026 20:50:00 GMTProductAIhttps://tariksenkal.com/en/posts/opentelemetry-genai-tracing/https://tariksenkal.com/en/posts/opentelemetry-genai-tracing/The gen_ai.* conventions are still unstable, and they are still the only standard for agent tracing. The span tree, operation names, MCP propagation, and version pinning.Thu, 13 Aug 2026 20:49:00 GMTObservabilityAIhttps://tariksenkal.com/en/posts/prompt-regression-testing/https://tariksenkal.com/en/posts/prompt-regression-testing/Build a golden set from production logs, wire a promptfoo CI gate, calibrate an LLM judge, and avoid the string-matching traps that break non-English evals.Thu, 13 Aug 2026 20:48:00 GMTEvaluationAIhttps://tariksenkal.com/en/posts/reasoning-token-economics/https://tariksenkal.com/en/posts/reasoning-token-economics/You pay for hidden thinking tokens. Effort levels replaced thinking budgets, task budgets arrived, caching breaks when you change them, and more thinking can hurt.Thu, 13 Aug 2026 20:47:00 GMTCostAIhttps://tariksenkal.com/en/posts/agent-sandboxing-permissions/https://tariksenkal.com/en/posts/agent-sandboxing-permissions/The lethal trifecta, the rule of two, a tool permission matrix, container isolation, egress filtering and an audit-log schema that doubles as a compliance trail.Thu, 13 Aug 2026 20:46:00 GMTSecurityAIhttps://tariksenkal.com/en/posts/agent-memory-systems/https://tariksenkal.com/en/posts/agent-memory-systems/How agent memory actually works: episodic, semantic and procedural stores, a real pgvector schema, the tool landscape, and controls against memory poisoning.Thu, 13 Aug 2026 20:45:00 GMTAgentsAIhttps://tariksenkal.com/en/posts/agent-harness-design/https://tariksenkal.com/en/posts/agent-harness-design/Why long-running agents wrap up early: context anxiety, why resets beat compaction, the generator-evaluator split and file-based handoff between agents.Thu, 13 Aug 2026 20:44:00 GMTArchitectureAIhttps://tariksenkal.com/en/posts/agents-in-production/https://tariksenkal.com/en/posts/agents-in-production/What it takes to run an AI agent in production: timeouts, idempotency, circuit breakers, prompt versioning, canary rollout and cost per successful task — with code.Thu, 13 Aug 2026 20:43:00 GMTAgentsAIhttps://tariksenkal.com/en/posts/enterprise-llm-budgeting/https://tariksenkal.com/en/posts/enterprise-llm-budgeting/A measured token-overhead figure, cache economics and FX exposure combined into a working unit-cost model — and an honest API-versus-self-host break-even.Thu, 13 Aug 2026 20:42:00 GMTCostAIhttps://tariksenkal.com/en/posts/mcp-server-security/https://tariksenkal.com/en/posts/mcp-server-security/Tool poisoning, rug-pull redefinition, confused deputy and cross-server shadowing: the MCP-specific attack surface, plus a working tool-definition pinning check.Thu, 13 Aug 2026 20:41:00 GMTSecurityAIhttps://tariksenkal.com/en/posts/owasp-agentic-top-10/https://tariksenkal.com/en/posts/owasp-agentic-top-10/Turning each item of the OWASP Agentic Top 10 (2026) into an implementable control — a config, a policy and a test — instead of restating the risk.Thu, 13 Aug 2026 20:40:00 GMTSecurityAIhttps://tariksenkal.com/en/posts/turkish-language-models-2026/https://tariksenkal.com/en/posts/turkish-language-models-2026/The Turkish open-weight LLM landscape with a Hugging Face link behind every claim: Cosmos, Kumru, Trendyol, Kanarya, BILGE — and how to spot models that do not exist.Thu, 13 Aug 2026 20:39:00 GMTModelsAIhttps://tariksenkal.com/en/posts/long-context-vs-rag/https://tariksenkal.com/en/posts/long-context-vs-rag/Does a 1M-token window make RAG obsolete? Context rot, the measured token cost of non-English text, and a third option: pushing long context into filesystem tool calls.Thu, 13 Aug 2026 20:38:00 GMTArchitectureAIhttps://tariksenkal.com/en/posts/computer-use-agents/https://tariksenkal.com/en/posts/computer-use-agents/How computer-use agents actually work: the 83.5% vs 20.6% benchmark gap OSWorld 2.0 exposed, the perceive-ground-act loop, grounding traps and the security model.Thu, 13 Aug 2026 20:37:00 GMTAgentsAIhttps://tariksenkal.com/en/posts/mcp-stateless-architecture-2026/https://tariksenkal.com/en/posts/mcp-stateless-architecture-2026/The 2026-07-28 MCP revision removed sessions, the Mcp-Session-Id header and the initialize handshake. MRTR, server/discover and a concrete migration plan.Thu, 13 Aug 2026 20:36:00 GMTToolsAIhttps://tariksenkal.com/en/posts/context-engineering/https://tariksenkal.com/en/posts/context-engineering/Context engineering is the discipline of deciding which tokens earn a place in a limited context window at each step. Context rot, lost-in-the-middle, pruning and summarisation strategies.Sun, 09 Aug 2026 20:35:00 GMTArchitectureAIhttps://tariksenkal.com/en/posts/autonomous-agent-architectures/https://tariksenkal.com/en/posts/autonomous-agent-architectures/Agents are more than a single loop. ReAct, plan-and-execute, supervisor-worker patterns, heterogeneous model use, and how to keep cost under control across agent fleets.Sun, 09 Aug 2026 20:34:00 GMTAgentsAIhttps://tariksenkal.com/en/posts/small-language-models/https://tariksenkal.com/en/posts/small-language-models/Small language models can beat large ones on high-frequency, narrow tasks. Latency, cost, on-device execution, fine-tuning and the realistic limits.Sun, 09 Aug 2026 20:33:00 GMTModelsAIhttps://tariksenkal.com/en/posts/model-routing-cascades/https://tariksenkal.com/en/posts/model-routing-cascades/Routing sends each request to the model whose cost and capability fit it. Cascade architectures, confidence thresholds, escalation paths and measuring the real cost-quality curve.Sun, 09 Aug 2026 20:32:00 GMTArchitectureAIhttps://tariksenkal.com/en/posts/reasoning-models-test-time-compute/https://tariksenkal.com/en/posts/reasoning-models-test-time-compute/Reasoning models generate thinking steps before answering. What test-time compute buys you, when it is wasted, and how it changes latency and cost.Sun, 09 Aug 2026 20:31:00 GMTModelsAIhttps://tariksenkal.com/en/posts/model-context-protocol/https://tariksenkal.com/en/posts/model-context-protocol/MCP gives models a standard interface to tools and data sources. The client-server model, the N-times-M problem, security boundaries and practical use.Sun, 09 Aug 2026 20:30:00 GMTToolingAIhttps://tariksenkal.com/en/posts/agent-evaluation/https://tariksenkal.com/en/posts/agent-evaluation/Evaluating an agent is not the same as scoring one answer. Task success, step quality, tool choice, cost and robustness testing in a practical agent evaluation framework.Sun, 09 Aug 2026 20:29:00 GMTAgentsAIhttps://tariksenkal.com/en/posts/guardrails-content-moderation/https://tariksenkal.com/en/posts/guardrails-content-moderation/Guardrails are the layer that constrains a model input and output with rules. Input filters, schema validation, output checks, grounding verification and the false-positive trade-off.Sun, 09 Aug 2026 20:28:00 GMTSafetyAIhttps://tariksenkal.com/en/posts/generative-engine-optimization/https://tariksenkal.com/en/posts/generative-engine-optimization/Generative search engines answer directly, so citations matter more than clicks. Practical ways to structure content so ChatGPT, Perplexity and AI Overviews quote it.Sun, 09 Aug 2026 20:27:00 GMTProductAIhttps://tariksenkal.com/en/posts/ai-in-legal-practice/https://tariksenkal.com/en/posts/ai-in-legal-practice/AI in law demands verifiable accuracy. Case-law search, grounding, mandatory citation, confidentiality, and why final responsibility stays with the lawyer.Sun, 09 Aug 2026 20:26:00 GMTApplicationsAIhttps://tariksenkal.com/en/posts/multimodal-models/https://tariksenkal.com/en/posts/multimodal-models/How do multimodal models combine image, text, and audio in a single system? We explain vision-language models, the shared embedding space, and real use cases with intuitive analogies and short code.Sun, 09 Aug 2026 20:25:00 GMTMultimodalAIhttps://tariksenkal.com/en/posts/diffusion-models-image-generation/https://tariksenkal.com/en/posts/diffusion-models-image-generation/How do diffusion models generate images? The forward and reverse processes, the path from noise to image, and Stable Diffusion's latent space — explained intuitively and accurately with everyday analogies.Sun, 09 Aug 2026 20:24:00 GMTGenerativeAIhttps://tariksenkal.com/en/posts/knowledge-distillation/https://tariksenkal.com/en/posts/knowledge-distillation/Knowledge distillation transfers a large teacher model’s knowledge into a small, fast student model. We explain soft labels, temperature, and the distillation loss, and why, how, and when it works, with intuitive examples.Sun, 09 Aug 2026 20:23:00 GMTEfficiencyAIhttps://tariksenkal.com/en/posts/lora-peft-fine-tuning/https://tariksenkal.com/en/posts/lora-peft-fine-tuning/An intuitive guide to LoRA and PEFT: adapting a model without training all of it, using small low-rank adapters, the memory/cost advantage, and practical usage tips.Sun, 09 Aug 2026 20:22:00 GMTFine-tuningAIhttps://tariksenkal.com/en/posts/prompt-injection-jailbreaks/https://tariksenkal.com/en/posts/prompt-injection-jailbreaks/What are prompt injection and jailbreaks, and how do they work? We explain the risks in RAG and agent systems, plus layered defenses, with plain everyday analogies.Sun, 09 Aug 2026 20:21:00 GMTSecurityAIhttps://tariksenkal.com/en/posts/function-calling-structured-output/https://tariksenkal.com/en/posts/function-calling-structured-output/Function calling and structured output turn a language model from a free-text chatter into a component you can wire into your systems with confidence. We explain JSON generation, schema enforcement, and robust integration practices with intuitive examples.Sun, 09 Aug 2026 20:20:00 GMTIntegrationAIhttps://tariksenkal.com/en/posts/llm-cost-optimization/https://tariksenkal.com/en/posts/llm-cost-optimization/Practical levers for lowering your LLM bill: reducing token consumption, prompt caching, matching the task to the model, batch processing, and the small model + RAG balance. Cost optimization in the right order, with intuitive analogies and example code.Sun, 09 Aug 2026 20:19:00 GMTCostAIhttps://tariksenkal.com/en/posts/rlhf-model-alignment/https://tariksenkal.com/en/posts/rlhf-model-alignment/Why and how do we make a model "helpful and safe"? We explain the three steps of RLHF, the reward model, and the simpler alternative DPO using everyday analogies.Sun, 09 Aug 2026 20:18:00 GMTAlignmentAIhttps://tariksenkal.com/en/posts/llm-observability/https://tariksenkal.com/en/posts/llm-observability/A plain-language guide to LLM observability: tracing, logging, evaluation loops, and how to protect quality and catch regressions early in production.Sun, 09 Aug 2026 20:17:00 GMTObservabilityAIhttps://tariksenkal.com/en/posts/ai-product-ux/https://tariksenkal.com/en/posts/ai-product-ux/How do you design user experience for AI products? A practical guide to managing uncertainty, citing sources, feedback, graceful failure, and building trust.Sun, 09 Aug 2026 20:16:00 GMTProductAIhttps://tariksenkal.com/en/posts/how-llms-work/https://tariksenkal.com/en/posts/how-llms-work/How do LLMs work? An intuitive yet technical guide to the Transformer architecture: next-word prediction, the attention mechanism, and layers explained simply.Sun, 09 Aug 2026 20:15:00 GMTLLMAIhttps://tariksenkal.com/en/posts/tokens-embeddings/https://tariksenkal.com/en/posts/tokens-embeddings/An intuitive guide to tokenization, word pieces, and embedding vectors: how the meaning of text turns into numbers and coordinates inside a language model.Sun, 09 Aug 2026 20:14:00 GMTFundamentalsAIhttps://tariksenkal.com/en/posts/what-is-rag-how-to-build/https://tariksenkal.com/en/posts/what-is-rag-how-to-build/What is RAG and how does it work? A practical, end-to-end architecture guide to document, chunk, embedding, vector search and context-grounded generation, with small code examples.Sun, 09 Aug 2026 20:13:00 GMTRAGAIhttps://tariksenkal.com/en/posts/vector-databases-guide/https://tariksenkal.com/en/posts/vector-databases-guide/An intuitive guide to vector databases: similarity search, ANN and the HNSW algorithm, a FAISS vs pgvector vs Qdrant comparison, and when to choose each one.Sun, 09 Aug 2026 20:12:00 GMTInfrastructureAIhttps://tariksenkal.com/en/posts/prompt-engineering-techniques/https://tariksenkal.com/en/posts/prompt-engineering-techniques/Clear instructions, few-shot, step-by-step reasoning, role assignment and output formatting: practical techniques to get better results from language models, with examples.Sun, 09 Aug 2026 20:11:00 GMTPracticeAIhttps://tariksenkal.com/en/posts/fine-tuning-vs-rag-vs-prompt/https://tariksenkal.com/en/posts/fine-tuning-vs-rag-vs-prompt/We compare fine-tuning, RAG, and prompting on cost, freshness, and accuracy. Which one should you pick and when? Includes a practical decision table.Sun, 09 Aug 2026 20:10:00 GMTDecisionAIhttps://tariksenkal.com/en/posts/reducing-llm-hallucinations/https://tariksenkal.com/en/posts/reducing-llm-hallucinations/Why do LLMs hallucinate and how can you reduce it? Practical measures with grounding, RAG, verification, self-checking, and temperature tuning.Sun, 09 Aug 2026 20:09:00 GMTReliabilityAIhttps://tariksenkal.com/en/posts/open-source-llms/https://tariksenkal.com/en/posts/open-source-llms/We compare open-source LLMs like Llama, Mistral, Qwen and Gemma against closed models: privacy, cost, fine-tuning freedom, and when each one wins.Sun, 09 Aug 2026 20:08:00 GMTEcosystemAIhttps://tariksenkal.com/en/posts/llm-evaluation/https://tariksenkal.com/en/posts/llm-evaluation/A guide to LLM evaluation: automatic metrics, LLM-as-judge, human evaluation, domain-specific test sets, and measuring faithfulness for RAG systems.Sun, 09 Aug 2026 20:07:00 GMTEvaluationAIhttps://tariksenkal.com/en/posts/mlops-ai-to-production/https://tariksenkal.com/en/posts/mlops-ai-to-production/What is MLOps? How to take an AI model to production: versioning, monitoring, latency/cost, the feedback loop, and safe deployment, explained in plain language.Sun, 09 Aug 2026 20:06:00 GMTMLOpsAIhttps://tariksenkal.com/en/posts/who-is-tarik-senkal/https://tariksenkal.com/en/posts/who-is-tarik-senkal/Who is İsmail Tarık Şenkal? The story of the co-founder of Hatson Teknoloji A.Ş. — a developer and entrepreneur working on Turkish artificial intelligence, large language models, and legal technologies.Sun, 09 Aug 2026 20:05:00 GMTProfileAILegal Techhttps://tariksenkal.com/en/posts/what-is-hatsontech/https://tariksenkal.com/en/posts/what-is-hatsontech/How did HatsonTech come about? The origin story of a Turkish AI startup founded in 2024 at ODTÜ Teknokent in Ankara: what problem it set out to solve, and why it builds its own products.Sun, 09 Aug 2026 20:04:00 GMTCompanyAILegal Tech