JevMade Sign in
← Back to experiments

Apps & data pipelines

laya-jev-GraphRAG

An agentic GraphRAG engine where every graph operation is a typed decision from a swappable System One model, and the LLM runs only once to write the answer.

Source screenshot of laya-jev-GraphRAG
SOURCE SCREENSHOTFull screenshot ↗

What it does

One environment variable swaps the whole decision layer between local Laya and cloud Jev, and an ablation mode runs both and logs the difference.

How you can use it

For a project that answers questions from documents, borrow the order of work: check which facts seem related, choose useful passages, then ask a writing AI for the answer. Start with the project's four stages. You can compare checks made on your computer with checks sent to an outside service. Neither set of AI checks guarantees that the answer is true.

Maker-reported (not independently measured by JevMade): Laya (local CUDA): ~33 ms/call, ~1.2 GB VRAM; Jev (cloud): ~50 ms/call · the generative LLM (Llama-3.1-8B) runs once at the very end

Primitives
choice, score, noul
Platform
Python
Added
Project created

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.