Coarse-to-fine retrieval over hierarchical knowledge graphs using Matryoshka embeddings. Mimics biological visual attention: peripheral (64d) -> parafoveal (128d) -> foveal (256d).
Two-stage e-commerce search: Matryoshka fine-tuning + BM25/SPLADE/dense hybrid retrieval + cross-encoder reranking. Recall@200 = 0.81, nDCG@20 = 0.54 on Amazon ESCI.
Local AI workbench for embeddings, summarization, and OpenAI Agent SDK–compatible workflows. Supports Gemma models, GPT-OSS tool-calling, hardware acceleration, caching, and rate limiting, plus cloud-offloaded, persona-driven summarization through Gemini.
An AI search system that turns a recruiter's plain-English request ("Senior ML engineer, Python, NLP, Bay Area") into ranked, explained candidate matches - built to replicate the retrieval architecture LinkedIn described in its own engineering blog, with measured results.
Add a description, image, and links to the matryoshka-embeddings topic page so that developers can more easily learn about it.
To associate your repository with the matryoshka-embeddings topic, visit your repo's landing page and select "manage topics."