Technical notes and project retrospectives on AI systems.https://lamisxdhia.pages.dev/en-ushttps://lamisxdhia.pages.dev/en/blog/whisper-tts-perceived-latency/https://lamisxdhia.pages.dev/en/blog/whisper-tts-perceived-latency/While building Sentinelle's voice copilot, the real latency problem wasn't where I expected it — and the fix cost nothing in audio quality.Mon, 03 Aug 2026 00:00:00 GMThttps://lamisxdhia.pages.dev/en/blog/rag-i-dont-know/https://lamisxdhia.pages.dev/en/blog/rag-i-dont-know/Why most RAG pipelines hallucinate out of overconfidence, and three concrete levers to make an agent recognise the limits of what it actually found.Tue, 02 Jun 2026 00:00:00 GMThttps://lamisxdhia.pages.dev/en/blog/langgraph-when-to-orchestrate/https://lamisxdhia.pages.dev/en/blog/langgraph-when-to-orchestrate/A linear chain is enough for a lot of LLM use cases. Here's the concrete signal that made me move some projects to LangGraph — and the ones where a simple chain was the right call.Tue, 14 Apr 2026 00:00:00 GMT