关于 AI、构建全球平台,以及按我们认可的标准交付软件的运维实务笔记。
Every company is choosing between building AI features and buying them. The decision rests on three questions, and the cost that gets underestimated is maintenance.
Finding a supplier is not a search problem. It is a resolution problem — deciding whether two descriptions refer to the same product, and whether a company can actually deliver it.
AI projects stall on data quality more often than on model choice. The specific defects in CRM and ERP records, and what to fix before adding AI.
Standard software is tested by assertion. AI output has no single correct answer. Here is how evaluation is actually built, and why most teams skip it.
The hard part of enterprise AI is not the model. It is giving it access to your CRM, ERP and documents without rebuilding an integration for every system.
Turkish data protection law was written before modern AI. What that means in practice for embeddings, prompt logs, model training and automated decisions.
Agent demos work and production deployments stall. The failure modes are predictable — error compounding, context limits, tool design, and no way to observe what happened.
Technoparks, tax exemptions, a national AI plan, new cloud infrastructure and a Tech Visa. What Türkiye offers a technology company, and what it does not.
AWS opened a Local Zone in Istanbul in May 2026. What a Local Zone is, how it differs from a full Region, and what it changes for builds in Türkiye.
HS classification looks like a text matching task and is not. Here is what the legal structure demands, and how to design an AI system around it.
Why a search system that works in English fails in Turkish, Russian or Chinese, and the retrieval design choices that actually fix it.
Retrieval and fine-tuning solve different failures. Diagnose which one you have before choosing, or you will spend months fixing the wrong problem.
Three tenant isolation models, what each one costs you operationally, and why adding AI to a SaaS product changes the calculation.
GITEX arrived in Istanbul this week for its first Türkiye edition. Notes from two days at the Istanbul Expo Center, and why this country is becoming a serious place to build AI software.
Connecting an application to a model API is a feature. AI platform engineering is the layer underneath it: retrieval, evaluation, versioning, fallback and cost control.
AI matching in B2B is usually keyword search with a better interface. This is what a real trade matching system computes: product fit, trade viability and counterparty fit.
Most B2B supplier verification still runs on emailed PDFs and photographed licences. Here is why that fails structurally, and what a real verification layer checks.