https://www.llm-works.ai/blog/ Technical writing on AI infrastructure, agent development, and fine-tuning. en-us Fri, 12 Jun 2026 04:36:43 +0000 https://www.llm-works.ai/blog/introducing-llm-news/ https://www.llm-works.ai/blog/introducing-llm-news/ Tue, 31 Mar 2026 00:00:00 +0000 An AI news curation service that monitors 50+ sources, summarizes and rates articles, clusters them into themes, and produces daily digests — all running on a local 14B model. https://www.llm-works.ai/blog/saia-verbs-for-llm-agents/ https://www.llm-works.ai/blog/saia-verbs-for-llm-agents/ Fri, 27 Mar 2026 00:00:00 +0000 Agents are code that interacts with LLMs. LLMs speak free-form text. How should this communication be structured? Current approaches, trade-offs, and what we're building. https://www.llm-works.ai/blog/building-agent-infrastructure/ https://www.llm-works.ai/blog/building-agent-infrastructure/ Thu, 19 Mar 2026 00:00:00 +0000 Agents today don't learn from experience. The missing piece isn't better models—it's the integration layer that connects inference, training, and deployment into a learning loop. https://www.llm-works.ai/blog/fine-tuned-3b-beats-haiku/ https://www.llm-works.ai/blog/fine-tuned-3b-beats-haiku/ Thu, 12 Mar 2026 00:00:00 +0000 Fine-tuned Qwen 3B outperforms Claude Haiku on constrained generation. Full scaling curve from 0.5B to 72B with SFT+DPO shows the quality cliff.