SciPhi

Interested in implementing RAG across document groups for your users? This cookbook demonstrates how to leverage R2R's built-in group permissions to quickly implement this. The cookbook covers creating and managing groups, adding users and documents to groups, and controlling access using R2R's group permissioning system - https://lnkd.in/e9C4c_XR

RAG & Businesses are just a match made in heaven. 💑 Recent algorithmic advancements like replacing positional encoding with ALiBi, Sparse attention and Flash Attention-2 have really extended allowable context-windows, but despite this - all the predictions of these techniques being RAG killers have turned out to be wrong. 📌 When making my most recent RAG application I stumbled onto a great open source RAG engine named R2R from SciPhi, that I found to be quite incredible. R2R seriously made life much easier for me as a developer to build, observe, and optimize RAG. 📌 Another feature that I really liked about R2R was its built-in user permissions and document management. I can see how this will make my next user facing application much easier to build. 📌 You can think of R2R as the Supabase for RAG – a complete platform that's bridging the gap between experimenting and deploying production-ready RAG applications. It even supports open-source models for my local-RAG. 📌 Part of what makes R2R easy to use is the fact that it is built around a simple RESTful API. This means I was able to deploy my application without digging into the nitty gritty details. It's also packing some serious heat with features like - Multimodal support (hello, .txt to .mp3!), - Hybrid search that combines semantic and keyword approaches, - implementing an advanced RAG technique HyDE (Hypothetical Document Embedding) and - Automatic knowledge graph generation. -------- 📌 R2R is an engine for building user-facing Retrieval-Augmented Generation (RAG) applications. It gives developers configurable vector search and RAG right out of the box, as well as direct method calls instead of the client-server architecture seen throughout the docs. The founder's vision is clear: create a tool that accelerates serious LLM application development without the pain points of existing open-source projects. R2R aims to be more opinionated about abstractions and integrations, resulting in a simple yet powerful tool. Whether you're diving into user management, craving some serious observability, or looking to extend your RAG capabilities, R2R's got your back. And for those of us who love a good UI, there's an open-source React+Next.js front-end to play with. --------- 📌 So the core workflow of R2R is the following, and you get all of these both out-of-the box, and also with a Python & Javascript SDKs so we can fully integrate it with your own codebase. - Ingest files into your Postgres vector database - Search over ingested files - Create a RAG (Retrieval-Augmented Generation) response - Perform basic user auth - Observe and analyze a deployed RAG engine. ------ 👉 Full Video - https://lnkd.in/g-FnfMa5 👉 R2R Official Doc - https://lnkd.in/gqRdw6QZ 👉 Github repo of R2R - https://lnkd.in/ggY37Wxs

Some major upgrades to the SciPhi cloud are now live 1.) Free RAG completions through gpt-4o-mini (w/ documented support for other providers like anthropic, openai, etc.) 2.) Fast document ingestion 3.) Agentic RAG 4.) Control and customizability - bring your own docker w/ R2R

The open source LLM community is really cooking now. We can run assistants directly from our laptops thanks to llama3.1's reliable tool calling. This quick video shows how to do this with R2R's new assistant API. Looking forward to seeing what people build, happy weekend hacking!

Thanks Mervin Praison ✅ for taking the time to make this awesome video about Triplex+R2R, it's super exciting to see the open source community dive into the new model!

Today SciPhi is open-sourcing Triplex, a SOTA LLM for knowledge graph construction. Why should you care about knowledge graphs? Take the following question: "What YC founders worked at Google?" Existing RAG methods cannot reliably answer queries like this, and that's where knowledges graphs (KGs) come in, like Microsoft's recent Graph RAG which shows great promise. However, only large foundation are capable of doing the extractions necessary to build these graphs, and so it can cost millions to create KGs from very large datasets. We built Triplex to solve this problem by making it 98% less expensive to build a knowledge graph. You can read more about how we created the Triplex on our blog here, https://lnkd.in/erQs9nsS, and you can try out a live demo right here - https://kg.sciphi.ai/. Triplex is so small that it can be used with SciPhi's R2R to build knowledge graphs, powered by Neo4j ,directly on your laptop. Triplex outperforms few-shot prompted gpt-4o at 1/60th the inference cost. Triplex is on HuggingFace here - https://lnkd.in/e463Pzdj and Ollama here - https://lnkd.in/euVD-4Bx Lastly, take a look at R2R (https://lnkd.in/eVNj2YfC) If you are interested in leveraging Triplex right away - the R2R cookbook will help you get started here - https://lnkd.in/eJ3XBxnK. Thanks to Nolan Tremelling and Shreyas Pimpalgaonkar for their amazing efforts in building this model.