https://gianlucatessitore.com/blog Notes on native LLMs, enterprise AI architecture and intelligent systems. en Wed, 29 Jul 2026 15:02:44 GMT https://gianlucatessitore.com/blog/the-craft-of-trustworthy-systems https://gianlucatessitore.com/blog/the-craft-of-trustworthy-systems Thu, 22 May 2025 09:00:00 GMT Philosophy Trust isn't a layer you bolt on. It's the result of clear domain models, conservative defaults, legible failure modes and an engineering culture that prefers honesty over magic. https://gianlucatessitore.com/blog/fine-tuning-again https://gianlucatessitore.com/blog/fine-tuning-again Sat, 10 May 2025 09:00:00 GMT Native AI When prompting plateaus and RAG isn't enough: a measured take on small, targeted fine-tunes — data curation, evaluation discipline and total cost of ownership. https://gianlucatessitore.com/blog/evaluating-llms-you-actually-ship https://gianlucatessitore.com/blog/evaluating-llms-you-actually-ship Wed, 02 Apr 2025 09:00:00 GMT Native AI Most public benchmarks don't tell you whether a model will survive your domain. A pragmatic harness for hallucination, safety, latency and task-specific accuracy — the metrics that decide deployments. https://gianlucatessitore.com/blog/agentic-workflows-without-the-chaos https://gianlucatessitore.com/blog/agentic-workflows-without-the-chaos Fri, 14 Mar 2025 09:00:00 GMT Agentic AI Designing agent graphs that scale: when to use tools vs. functions, how to bound autonomy, observability primitives, and the guardrails that turn demos into production. https://gianlucatessitore.com/blog/rag-is-not-a-product https://gianlucatessitore.com/blog/rag-is-not-a-product Fri, 21 Feb 2025 09:00:00 GMT Architecture Retrieval-augmented generation is plumbing. The product is the system around it: chunking strategy, evaluation, freshness, fallbacks and the quiet work of keeping it accurate. https://gianlucatessitore.com/blog/salesforce-as-an-ai-substrate https://gianlucatessitore.com/blog/salesforce-as-an-ai-substrate Sat, 18 Jan 2025 09:00:00 GMT Salesforce Why the platform — Data Cloud, Agentforce, Einstein and Flow — is one of the most underrated AI runtimes in the enterprise, and how to architect on it without losing portability.