Build this very cool personal notch buddy. Should I publish it ? @NerdsRoom_ @cognee_ @Paytm

Jev is great at zero-shot classification, but specialist classifiers will dominate commercial use cases. @trycua tuned a tiny model that scored 99.7% on their form-filling eval. Hosted Jev scored 83.6%. I tuned GLiNER 2.5 on a task in 51 minutes yesterday and it crushes Jev. And it's local. And 8.8x faster: You too can do this. Linked post in comments.

I've been at three search-focused events in the past three days in Berlin, so here's a recap of everything that happened: - Wednesday, @qdrant_engine and @cognee_ Bring Your Own Demo night: such a great energy and turnout from the community. People are building genuinely interesting things with Qdrant, and it's great to see how also less-technical users can easily get started with Vector Search using Qdrant and coding agents :) I talked about how you can level up your production search using Qdrant Cloud for server-side embedding inference and GPU indexing, if it sounds interesting hit me up and we can talk :)) - Thursday, Future of Search: my talk here was titled "Vector Search for Everyone" and centered around how you can use Qdrant to quickly get started with your internal/side project vector search experimentation, and you can scale that with the same client interface from a self-hosted Docker container to a full-blown distributed deployment. The main point that emerged from the discussion afterwards is how much we need frameworks to validate and evaluate search pipelines to justify an investment in them, and I feel like that's definitely worth focusing efforts on. - Thursday, Vector Space Stream: a 4-hrs, reseaech-focused stream in which we shared all the amazing research we're doing at Qdrant, from robotics and edge to scaling cloud inference, to separating computing and storage for serverless inference. With my colleague Jonas and Ivan, we talked about TurboQuant, from the theory behind it to how you can use it in Qdrant to save $$ on storage and memory without losing too much recall. If you weren't able to catch it, here's the recording of the whole stream: piped.video/live/PxGlBlqTxJI… You can find all the slides at: qdrant-talks.clelia.dev All in all it was in incredibly dense week, and I'm super grateful to all my teammates for the support and help they gave me, and all the people I met for the things I learnt from them🙏 See you next month, Berlin✨

Reminder to join our webinar on how to build a company brain, Tuesday 22 September. 45 min, online and free. Which data sources to connect on day one, and what your agents can ask the graph once it's built. 20:00 CEST / 11:00 PT luma.com/cognee-flfx

Memory related posts keep going viral because people can actually feel the problem! As a leader in the space building @cognee_, we wrote a detailed guide which gets deep into what agent memory actually is. Read more here: cognee.ai/agent-memory?utm_s…

Berlin to San Francisco is 9,108 km. My talk was 5 minutes long, in front of 20k people, for @BerkeleyRDI Xcelerator ! Worth it? 1000% Why? Because we believe there should be an alternative to Anthropic and OpenAI holding all your data.