Rnest.io — La mètis
Responsible for AI innovation and development for automated analysis of large corpora from web crawlers. Designed the agentic RAG system end to end:
Discovery loop — autonomous query derivation, hybrid dense/BM25 retrieval with RRF fusion, cross-encoder reranking, LLM-based document selection.
Inductive analysis & reports — inspired by Grounded Theory: insight extraction and typing, bottom-up grouping into “dynamics”, axial-coding chapter plans, citation-grounded drafting with multi-pass correction.
Analyst workbench — human-in-the-loop web app to steer each step, edit the analysis frame, add charts and maps, and export to Word and HTML.
Architecture — polylith, FastAPI, YAML-driven multilingual prompts, fully local LLM.