LlamaIndex newsletter 2023–10–24

Hello Llama Fans 🦙!

Welcome back to our newsletter covering new features, guides, integrations, webinars, tutorials, and more. Got a project, blog, or video you’re proud of? Let’s spotlight it! Contact us at news@llamaindex.ai.

Plus, for direct updates in your email, just head to our homepage and subscribe to our newsletter.

🤩 First, the highlights:

1. QueryFusionRetriever Launch: Inspired by Adrian Raudaschl’s RAG-Fusion, enhancing multiple query generation with LLMs. Tweet, Docs.
2. Router Fine-Tuning: Our innovative router fine-tuning approach has achieved an outstanding 99% match rate, outpacing both the gpt-3.5’s 65% and the base model’s 12%. Tweet, Docs.
3. Fusion Retriever Guide: Guide on building an advanced Fusion Retriever from scratch. Docs
4. Amazon Bedrock LLMs and AI21 Labs LLMs: We have expanded our LLM compatibility, now seamlessly integrating with both Amazon Bedrock and AI21 Labs models.

✨ Feature Releases and Enhancements:

• QueryFusionRetriever: We introduced the QueryFusionRetriever, inspired by Adrian Raudaschl’s work on RAG-Fusion. This retriever allows users to generate multiple queries with LLMs, run various retrieval methods, and apply reciprocal rank fusion for improved results. Tweet, Docs.
• Router Fine-Tuning: We introduced router fine-tuning (V0) for improved LLM automated decision-making. Our approach achieved a 99% match rate, outperforming gpt-3.5’s 65% and the base model’s 12%. Tweet, Docs.
• SQLRetriever: We introduce SQLRetriever, merging Text-to-SQL and RAG, enabling a RAG pipeline setup over SQL databases for structured table node retrieval and response synthesis. Tweet, Docs.

🗺️ Guides:

• Tutorial guide on Building an Advanced Fusion Retriever from Scratch.

✍️ Tutorials:

• Saurav Joshi’s tutorial on Complex Query Resolution through LlamaIndex Utilizing Recursive Retrieval, Document Agents, and Sub Question Query Decomposition.
• Greg Loughnane and Chris Alexiuk tutorial on tackling domain-specific fine tuning using LlamaIndex.
• Vishwas Gowda’s blog post on Streamlit LLM Hackathon winning app — FinSight using LlamaIndex.
• Emanuel Ferreira’s blog post on the RA-DIT paper and its implementation in LlamaIndex.
• Yujian Tang’s blog post on Chat with Towards Data Science using LlamaIndex.
• Sudarshan Koirala tutorial on Chat with documents with Pinecone and LlamaIndex.
• Sudarshan Koirala tutorial on Combined Text-TO-SQL + Semantic Search with LlamaIndex.
• PromptEngineer tutorial on building LLM-powered financial analyst with LlamaIndex.

⚙️ Integrations & Collaborations:

• Gradient AI: We introduce a collaboration with Gradient AI to easily integrate fine-tuned LLMs into your LlamaIndex RAG pipeline. Tweet, Blogpost.
• PrivateGPT: PrivateGPT partners with LlamaIndex allowing private document interactions using default or custom integrations. Tweet.
• VectorFlow & LlamaHub Collaboration: VectorFlow’s open-source vector-embedding pipeline now leverages LlamaHub for data connectors to streamline code and reduce maintenance. Tweet.
• Amazon Bedrock & AI21 Labs LLMs: We’ve broadened our LLM compatibility range by integrating with Amazon Bedrock LLMs and AI21 Labs LLMs.
• DashVector: We have introduced an integration with DashVector, a robust, fully-managed vectorDB service.
• Tencent Cloud: We’ve integrated with Tencent Cloud VectorDB.
• PGVectorStore within LlamaIndex has been enhanced to support custom Postgres schemas. This facilitates better index management and promotes easy schema-based versioning.
• We now accommodate custom models that align with the OpenAI-compatible API.

🎥 Webinars:

• Wenqi Glantz workshop webinar on Evaluation-Driven Development (EDD).
• Webinar showcasing the winning projects from the recent AGI House hackathon: “Build, Test, and Launch LLM Apps”. This event was co-sponsored by LlamaIndex, TruEra, and Pinecone.