Mayssa Ben Romdhane ยท Blog https://mayssa-portfolio.pages.dev/blog/ Articles by AI engineer Mayssa Ben Romdhane on building LLM systems for production: RAG pipelines, AI agents, LLM evaluation and computer vision. en Mon, 09 Nov 2026 20:29:21 GMT How to evaluate a RAG system without fooling yourself https://mayssa-portfolio.pages.dev/blog/how-to-evaluate-a-rag-system/ https://mayssa-portfolio.pages.dev/blog/how-to-evaluate-a-rag-system/ Wed, 14 Oct 2026 09:00:00 GMT A high retrieval score can still hide wrong answers. The reference sets, metrics and release gate I use to know whether a RAG change made answers better. RAG LLM evaluation Running open-weight LLMs on your own GPUs https://mayssa-portfolio.pages.dev/blog/open-weight-llms-on-your-own-gpus/ https://mayssa-portfolio.pages.dev/blog/open-weight-llms-on-your-own-gpus/ Tue, 22 Sep 2026 09:00:00 GMT When health data cannot leave the building, the model has to come to the data. What self-hosting open-weight LLMs costs, and what it buys you. LLM infrastructure RAG Human approval steps for LangGraph agents https://mayssa-portfolio.pages.dev/blog/human-approval-in-langgraph-agents/ https://mayssa-portfolio.pages.dev/blog/human-approval-in-langgraph-agents/ Thu, 03 Sep 2026 09:00:00 GMT An agent that sends reports or changes records should stop and ask first. How I build approval steps, retries and tracing into LangGraph workflows. AI agents Hybrid search: when BM25 beats embeddings https://mayssa-portfolio.pages.dev/blog/hybrid-search-bm25-and-embeddings/ https://mayssa-portfolio.pages.dev/blog/hybrid-search-bm25-and-embeddings/ Wed, 12 Aug 2026 09:00:00 GMT Embeddings miss exact codes, product names and section numbers that keyword search finds at once. Why my RAG systems combine BM25 and vector search. RAG Zero-shot road segmentation with foundation models https://mayssa-portfolio.pages.dev/blog/zero-shot-road-segmentation/ https://mayssa-portfolio.pages.dev/blog/zero-shot-road-segmentation/ Wed, 08 Jul 2026 09:00:00 GMT What foundation models like SAM and CLIP got right on road scenes without labelled training data, where they failed, and how we measured it at Continental. Computer vision Auto-annotating camera and LiDAR data with foundation models https://mayssa-portfolio.pages.dev/blog/auto-annotating-camera-and-lidar/ https://mayssa-portfolio.pages.dev/blog/auto-annotating-camera-and-lidar/ Wed, 17 Jun 2026 09:00:00 GMT Labelling camera and LiDAR data is slow and expensive. How foundation-model pre-labels and camera-LiDAR fusion speed it up without lowering quality. Computer vision