DEV Community

elvisyao007 profile picture

elvisyao007

AI implementation engineer. I make LLM systems reliable, measurable, and production-ready — eval-driven, secure on-prem RAG/agents. Building in public from one RTX 5090.

Joined Joined on  github website
A Chinese 8B model beat the Western 8B models at Japanese RAG. I still wouldn't put it in the default deployment — and that distinction is the point.
elvisyao007 profile

A Chinese 8B model beat the Western 8B models at Japanese RAG. I still wouldn't put it in the default deployment — and that distinction is the point.

#llm #rag #machinelearning #japan
Comments
4 min read

Want to connect with elvisyao007?

Create an account to connect with elvisyao007. You can also sign in below to proceed if you already have an account.

Create Account
Already have an account? Sign in
Which Chinese open-source parser is better for Japanese RAG? It's a crossover — BM25 says DeepDoc, dense says MinerU
elvisyao007 profile

Which Chinese open-source parser is better for Japanese RAG? It's a crossover — BM25 says DeepDoc, dense says MinerU

#rag #llm #machinelearning #japan
Comments
4 min read
Structured parsing helps dense retrieval more than it helps BM25 — measured on Japanese docs, and the gap doubled
elvisyao007 profile

Structured parsing helps dense retrieval more than it helps BM25 — measured on Japanese docs, and the gap doubled

#rag #llm #machinelearning #japan
Comments
4 min read
Half of agent evaluation needs no LLM judge — and it's the half that catches the failures that actually hurt
elvisyao007 profile

Half of agent evaluation needs no LLM judge — and it's the half that catches the failures that actually hurt

#ai #llm #agents #machinelearning
Comments
5 min read
Does a Chinese document parser actually work on Japanese PDFs? I measured it — and the answer is 'it depends on the font path'
elvisyao007 profile

Does a Chinese document parser actually work on Japanese PDFs? I measured it — and the answer is 'it depends on the font path'

#rag #llm #japan #machinelearning
Comments
5 min read
My local-LLM benchmark gave every model a perfect score. That was the most useful failure of the project.
elvisyao007 profile

My local-LLM benchmark gave every model a perfect score. That was the most useful failure of the project.

#llm #machinelearning #ai #japan
Comments
5 min read
I built a self-hosted LLM stack that grades itself — audit trail, per-user auth, and a built-in acceptance test
elvisyao007 profile

I built a self-hosted LLM stack that grades itself — audit trail, per-user auth, and a built-in acceptance test

#llm #selfhosted #devops #ai
Comments
6 min read
Your RAG dashboard can hide a failing retriever: detecting silent regression
elvisyao007 profile

Your RAG dashboard can hide a failing retriever: detecting silent regression

#rag #evaluation #llm #opensource
Comments
3 min read
I built a tiny tool to catch the metric trap from my last post
elvisyao007 profile

I built a tiny tool to catch the metric trap from my last post

#rag #evaluation #python #opensource
Comments
1 min read
The 33 'grounded-but-wrong' answers were a metric artifact: how ID-based context recall lies on multi-answer datasets
elvisyao007 profile

The 33 'grounded-but-wrong' answers were a metric artifact: how ID-based context recall lies on multi-answer datasets

#rag #evaluation #llm #retrieval
Comments
4 min read
faithfulness spread = 0.000: what self-grading RAG eval actually looks like
elvisyao007 profile

faithfulness spread = 0.000: what self-grading RAG eval actually looks like

#rag #llm #ai #machinelearning
Comments 1
4 min read
My RAG's faithfulness was 0.67. 1 in 3 answers were still wrong.
elvisyao007 profile

My RAG's faithfulness was 0.67. 1 in 3 answers were still wrong.

#rag #llm #ai #machinelearning
Comments 1
6 min read
loading...