Raj

Building -- Agents, Memory, Graph RAG

[Raj @raj]

The reranker, fusion and pre-ranking filters are live in Ragwalla now:

{"ranking_options": {"ragwalla": {"reranker": {"model": "typesafe/jev", "candidates": 100}}}}

Full write-up: https://t.co/NH6aMBlbWz

[Raj @raj]

I benchmarked Jev at two points of a RAG pipeline over 511,962 documents. Reranking a deep candidate pool raised the share of answer docs in the top 10 from 77.8% to 87.0%. Routing searches by app lost.

[Raj @raj]

Two places we want to try Jev next, both untested:

Scoring whether what a search found is enough to answer, and searching again if not.

Dropping low-fit passages instead of always passing 10, so the model sees fewer near-duplicates.

[Raj @raj]

Full post with charts: https://t.co/V59ZiM2LYQ

[Raj @raj]

I benchmarked @Cloudflare's Clairic against gpt-6-luna for entity extraction uses cases. The results are jarring.

[Raj @raj]

We are keeping gpt-6-luna, with a schema, as the extraction model for knowledge graphs. We are not building one over the benchmark corpus. Clairic mode stays in Ragwalla for anyone who wants a quick list of names from a document, but it is not the right tool for building a graph supported by entity extraction.