Hadi Mohammadi — Blog https://mohammadi.cv/blog/ Field notes on building and evaluating production LLM systems, and open guides for people studying AI and data science. en CC BY-SA 4.0, Hadi Mohammadi Fri, 07 Aug 2026 09:00:00 +0200 What an LLM judge actually does when you are not looking https://mohammadi.cv/blog/what-an-llm-judge-actually-does/ https://mohammadi.cv/blog/what-an-llm-judge-actually-does/ Fri, 07 Aug 2026 09:00:00 +0200 Three local LLM judges graded the same adversarial arithmetic benchmark in both presentation orders. One picked whichever answer came first 91% of the time; another graded effort and agreed with ground truth less often than a coin flip. Field notes Calling LLMs well https://mohammadi.cv/blog/calling-llms-well/ https://mohammadi.cv/blog/calling-llms-well/ Fri, 07 Aug 2026 09:00:00 +0200 Structured outputs, retries, streaming, caching and cost control — the unglamorous 80% of production LLM work. Modern AI Engineering Retrieval-augmented generation https://mohammadi.cv/blog/retrieval-augmented-generation/ https://mohammadi.cv/blog/retrieval-augmented-generation/ Fri, 07 Aug 2026 09:00:00 +0200 Chunking, hybrid search, reranking and query rewriting: RAG treated as an information-retrieval problem, and how it actually fails. Modern AI Engineering Agents and MCP https://mohammadi.cv/blog/agents-and-mcp/ https://mohammadi.cv/blog/agents-and-mcp/ Fri, 07 Aug 2026 09:00:00 +0200 Tool loops, the Model Context Protocol, planning patterns and agent evaluation — and when a plain workflow beats an agent. Modern AI Engineering Fine-tuning and preference optimization https://mohammadi.cv/blog/fine-tuning-and-preference-optimization/ https://mohammadi.cv/blog/fine-tuning-and-preference-optimization/ Fri, 07 Aug 2026 09:00:00 +0200 A decision table for RAG versus prompting versus LoRA versus DPO and GRPO, then the mechanics of each. Modern AI Engineering Evaluation and LLM-as-judge https://mohammadi.cv/blog/evaluation-and-llm-as-judge/ https://mohammadi.cv/blog/evaluation-and-llm-as-judge/ Fri, 07 Aug 2026 09:00:00 +0200 The five-level eval hierarchy, how to build an LLM judge you can defend, why agreement is the metric behind the metric. Modern AI Engineering Serving and inference optimization https://mohammadi.cv/blog/serving-and-inference-optimization/ https://mohammadi.cv/blog/serving-and-inference-optimization/ Fri, 07 Aug 2026 09:00:00 +0200 Prefill and decode, vLLM and continuous batching, KV caches, quantization, and the cost arithmetic behind API versus self-host. Modern AI Engineering