Woke up to 🚀 cognee hitting 5000 stars!

Thank you for the trust, feedback and code you pour in. Let’s keep building 🛠️

find us on @github.com trending!

Taken together, the results support the use of hyperparameter optimization as a routine part of deploying retrieval-augmented QA systems. Gains are possible and sometimes substantial, but they are also dependent on task design, metric selection, and evaluation procedure.

We evaluate on three established multi-hop QA benchmarks: HotPotQA, TwoWikiMultiHop, and Musique. Each configuration is scored using one of three metrics: exact match (EM), token-level F1, or correctness.

Why does AI memory matter?

LLMs can’t give us details about our data, they "forget" or simply don’t know the details.

Yesterday, we released our paper, "Optimizing the Interface Between Knowledge Graphs and LLMs for Complex Reasoning"

We have developed a new tool to enable AI memory optimization that considerably improve AI memory accuracy for AI Apps and Agents. Let’s dive into the details of our work 📚

Got featured in @neo4j.com ’s weekly wrap-up!

From the links below, you can find us chatting on GraphRAG + cognitive science (a small demo included)
+
fresh Neo4j news (Aura Analytics, NODES CFP, MCP hacks).

Don’t miss it out.

2k Github stars!
And, even more important, telemetry shows people running cognee in production!

We benchmarked cognee, Mem0 and Zep/Graphiti using HotPotQA! Cognee topped the charts, boosted further by Dreamify, our optimized pipeline 🚀. Graphiti shared their scores with us and we'll verify them soon!

Dive in the details 👇🏼

First time presenting our 89,4% accuracy on HotPot QA at a meetup and it was great.

We showed up as almost the whole @cognee.bsky.social team, answered great questions, met amazing people.

For the evals, go and check our repo 👇

Short read, big insights 👇

We took knowledge graphs up a notch with formal ontologies.

You’ll find how we merge domain expertise + graph structures for more detailed insights at
@cognee.bsky.social in the below blog. Check it out.

Want your AI to catch sneaky data errors and go beyond basic vector search?

We just released a blog on using knowledge graphs for smarter QA. Check it out!

ps: a sneak peek from the blog

Let's turn your data into graph/vector representation!
We will cognify it.
Inspired by #graphrag and #rag, we have a better approach!

Why?
- Increase retrieval accuracy by 30% (measured on hotpotqa)
- Get the data you need
- Reduce the number of LLM calls
- Extract the data from your own data

Here is our design of the evaluation system for #graphrag and #aimemory system that we built
It will be merged today

Curious how others approach #llm evals

entering bluesky with this amazing photo

just dropped off some cognee merch for the dlthub crew

let's make data processing great again!
@matthausk.bsky.social
@datateam.bsky.social