LlamaIndex Newsletter 2024-04-30

Greetings, LlamaIndex fans! 🦙

It’s delightful springtime weather out here in San Francisco and we hope you’re having a good day! Check out this week’s summary of news, guides and tutorials.

🤩 The highlights:

• Day 0 support for Microsoft’s Phi-3 Mini! Tweet
• create-llama now supports Llama 3 and Phi-3 and has lots of new features! Tweet
• Simon was on a security podcast! Tweet

✨ Feature Releases and Enhancements:

1. Jina AI released powerful new open-source rerankers and we have day 0 support as usual! Tweet
2. Phi-3 mini was released by Microsoft, a powerful new small model, and we put it through its paces (spoiler: it’s good!) and released day-0 support via Ollama! Tweet
3. Our create-llama application generator was updated with many features including being able to show the sources it retrieved from, as well as Llama3 and Phi-3 support. Build an app from scratch in 30 seconds! Tweet
4. Language Agent Tree Search (LATS) is a powerful new technique that iteratively plans out an array of potential futures, interleaving tool use and reflection to solve problems. We released a Llama Pack implementation. Tweet

🎥 Demos:

• Memary is a reference implementation of using long-term memory in knowledge graph form for building agents. Tweet
• Our hackathon winners wrote a blog post about their winning project, a knowledge-retrieval bot trained on documentation, including how they built it. Tweet

🗺️ Guides:

• Co-founder Jerry shared his latest deck, a guide to building a context-augmented research assistant that enables multi-hop Q&A, reflection and more. Slides, tweet
• Corrective RAG or CRAG adds a retrieval evaluation module that determines whether the retrieved context is “correct” and improves retrieval. Check out this guide on how to build it step-by-step! Tweet
• Jerry also went in-depth on the ingredients necessary for building a complex agent. Tweet
• Michael of KX Systems demonstrated making retrieval a multi-hop process for better results. Tweet
• A reference architecture for advanced RAG with LlamaIndex and AWS Bedrock. Tweet

✍️ Tutorials:

• Build a best-in-class RAG application using Qdrant as a vector store, Jina AI embeddings, and Mixtral 8x7b as the LLM. Tweet
• Learn 3+ patterns for building LLM apps on AWS with LlamaIndex. Tweet
• A 9-part series on taking RAG from prototype to production. Tweet

🎥 Webinars:

• KX Systems are hosting a webinar on May 1 about getting the most out of LlamaParse! Tweet
• Co-founder Simon appeared on the MLSecOps podcast talking about security in LLM applications. Tweet

👯‍♀️ Community:

We launched a LlamaIndex user group in Korea!