<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"><channel>
<title>BonFrame Engineering</title><link>https://www.bonframe.com/en/blog/</link>
<description>Local video AI, retrieval and batch editing engineering notes.</description>
<language>en</language>
<item><title>Selecting grounded video spans with an 8B local model using 81% fewer input tokens</title><link>https://www.bonframe.com/en/blog/grounded-video-rag-local-8b/</link><guid>https://www.bonframe.com/en/blog/grounded-video-rag-local-8b/</guid><pubDate>Mon, 24 Aug 2026 00:00:00 GMT</pubDate><description>How retrieval, transcript IDs and deterministic assembly reduced long-video generation context from about 4,600 to 860 tokens in a small locked holdout—and where the system still failed.</description></item>
<item><title>Grounded video RAG benchmark: protocol, results and claim boundaries</title><link>https://www.bonframe.com/en/research/grounded-video-rag-benchmark-method/</link><guid>https://www.bonframe.com/en/research/grounded-video-rag-benchmark-method/</guid><pubDate>Mon, 24 Aug 2026 00:00:00 GMT</pubDate><description>The locked eight-case protocol behind BonFrame&#039;s 3B, 8B and DeepSeek full-transcript comparison, including the earlier negative result and known failures.</description></item>
</channel></rss>