İsmail Tarık Şenkal

How I work

From idea to product, in three steps

1. 01

   Adapt

   I adapt a language model to the problem at hand and to Turkish.

2. 02

   Ground

   I connect it to real sources with RAG so it can’t “make things up.”

3. 03

   Ship

   I carry it into production and make it scalable and reliable.

Climate7 min read

What Is COP31 in Antalya? Dates, Hosting Deal and Key Issues

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.

Read more

• 
  Cost7 min

  Prefix Caching Economics: When Does the Cache Actually Pay?

  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.

• 
  Infrastructure7 min

  Speculative Decoding: Getting Faster Tokens From the Same Model

  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.

• 
  Efficiency7 min

  Quantization in Production: FP8, NVFP4, and Where Quality Actually Drops

  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.

• 
  Architecture8 min

  Compaction or Reset? Managing Context in Long Agent Sessions

  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.

• 
  Security9 min

  Defeating Prompt Injection by Design: CaMeL and Capability-Based Defence

  Better prompts do not solve prompt injection. CaMeL, the dual-LLM pattern, capability-based policy and the lethal trifecta — defences enforced outside the model.

• 
  Product8 min

  MCP Apps: Interactive UI Inside the Conversation

  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.

• 13 Aug 2026A2A: How Agents Talk to Each Other — and Why It Isn't MCP's Rival7 min
• 13 Aug 2026Multi-Agent Systems: 15× the Tokens, 90% Better — Which One Is You?6 min
• 13 Aug 2026Erasing Personal Data From a RAG Index — and From Fine-Tuned Weights6 min
• 13 Aug 2026Does llms.txt Actually Work? What the Data Says, and What v2 Changed6 min
• 13 Aug 2026OpenTelemetry GenAI: Tracing Agents on a Standard Schema5 min
• 13 Aug 2026Prompt Regression Testing: Putting Evals in CI6 min

Expertise

• Large Language Models

  I train and ship models fine-tuned for the Turkish legal language.

• Natural Language Processing

  Understanding Turkish and multilingual text: classification, inference and meaning-based processing.

• RAG

  Retrieval architectures that ground answers in real documents and stop the model from making things up.

• Semantic Search & Vector DB

  Search based on meaning rather than keywords — precise, fast retrieval across large archives.

• MCP & Agent Systems

  Agent architectures that use tools and route intelligently via the Model Context Protocol.

• Model Training & MLOps

  Training, fine-tuning and continuous deployment on large-scale infrastructure. TRUBA and HPC experience.

• Air Quality Modeling

  With an environmental-engineering background: modeling air pollutants and forecasting air quality.

Journey

1. Education

   Engineering and an interest in AI

   I deepened my interest in software and artificial intelligence through academic and applied work, focusing on natural language processing.

2. Research

   Turkish NLP and large-scale models

   I took part in large-scale AI work in environments such as TÜBİTAK events, TRUBA (the National e-Science infrastructure) and EUMaster4HPC.

3. 2024

   Founding HatsonTech

   I co-founded Hatson Teknoloji A.Ş. at METU Technopark in Ankara, with one goal: AI products focused on Turkish.

4. Today

   Legal language model

   I develop LLM and RAG architectures built for the Turkish legal language and run AI projects with public institutions.