LlamaIndex Newsletter 2024-06-11

Hello Llama Fans🦙

Step into this week's edition of the LlamaIndex newsletter, where we bring you a slew of exciting updates, in-depth guides, demos, enriching educational tutorials, and webinars designed to enhance your experience and understanding of our platforms and tools.

🤩 The highlights:

• Enhanced Memory Modules: New memory modules in LlamaIndex boost agentic RAG capabilities with Vector Memory for message storage and retrieval, and Simple Composable Memory for integrating multiple memory sources. Notebook1, Notebook2, Tweet.
• Create-llama and E2B Integration: Launched integration turns agents into advanced data analysts, enabling Python coding for data analysis and generating detailed files like graph images. Tweet.
• LlamaParse and Knowledge Graphs: Guide on integrating LlamaParse with Knowledge Graphs to develop RAG pipelines and agents for complex query handling.
• Prometheus-2 RAG Evaluation: Guide on using Prometheus-2, an affordable, transparent LLM based on Mistral models for effective RAG application evaluation with customized criteria.
• Agentic RAG : Video tutorial on Agentic RAG covering memory, planning, and reasoning, enhancing knowledge retrieval and agent capabilities.

✨ Feature Releases and Enhancements:

1. We have introduced new memory modules in LlamaIndex for enhancing agentic RAG capabilities. The Vector Memory module enables storage and retrieval of user messages using vector search, while the Simple Composable Memory module allows for integration of multiple memory sources. Notebook1, Notebook2, Tweet.
2. We have launched an integration between Create-llama and E2B’s sandbox, transforming agents into powerful data analysts. This new feature allows agents to write Python code for data analysis and return comprehensive files, like graph images, enhancing the scope of what agents can accomplish. Tweet.
3. We have launched an integration with Nomic-Embed-Vision that transforms Nomic-Embed-Text into a multimodal embedding that excels in handling image, text, and combined tasks, outperforming OpenAI CLIP with open access for all. Notebook.

🗺️ Guides:

• Guide to Integrating LlamaParse with Knowledge Graphs to develop a RAG pipeline for sophisticated query retrieval, and create an agent capable of answering complex queries effectively.
• Guide to Using Prometheus-2 for RAG Evaluation for assessing RAG applications, built on Mistral base models, it offers an affordable and transparent solution for evaluation, capable of direct assessments, pairwise rankings, and tailored criteria, ensuring alignment with human judgments.
• Guide to Three Forms of Query Rewriting for RAG to enhance RAG pipelines with techniques like sub-question decomposition, HyDE for aligning questions with embedding semantics, and step-back prompting for tackling complex queries more effectively.

🖥️ Demos:

• Laurie Voss’s LLM-powered file organizer project that categorizes files into folders based on LLM-decided categories without renaming them, ensuring important filenames remain intact. It organizes your files in multiple passes to balance folder sizes, resulting in descriptive yet practical folder names to help you find files easily.

✍️ Tutorials:

• Laurie Voss’s video tutorial on transitioning from basic RAG to fully agentic knowledge retrieval, featuring real-world code examples that cover routing, memory, planning, tool use, and advanced agentic reasoning methods like Chain of Thought and Tree of Thought, along with insights into observability, controllability, and customizability.
• Prince krampah's tutorials on Agentic RAG Systems, offering comprehensive insights into advanced system building with detailed explanations on router query engines, function calling, and multi-step reasoning across complex documents.
• kingzzm’s tutorial on Three Forms of Query Rewriting for RAG to enhance RAG pipelines with techniques like sub-question decomposition, HyDE for aligning questions with embedding semantics, and step-back prompting for tackling complex queries more effectively.
• Rajdeep Borgohain's tutorial to build a customer-support voicebot with advanced features like speech-to-text and text-to-speech, integrated into a RAG pipeline for efficient handling of customer support exchanges using Inferless, LlamaIndex, faster-whisper, Piper, and Pinecone.
• Pavan Mantha's tutorial on securing RAG apps using Azure for application security, including identity management, secure key storage, and managed Qdrant.

📹 Webinar:

• Join us for our webinar with Tomaz Bratanic from Neo4j on LlamaIndex property graph for insights into high-level and low-level graph construction, retrieval, and knowledge graph agents.