• Weaviate Newsletter
  • Posts
  • RAG Everywhere: Late Chunking, Verba Updates, Advanced Strategies, and More

RAG Everywhere: Late Chunking, Verba Updates, Advanced Strategies, and More

Author

Femke Plantinga
September 12, 2024

Hello Weaviate Community, 🤗

This week, we're diving into late chunking for better context retention, our Q3 Community Survey, upcoming events, and introducing new team members.

Let's get started!

Late Chunking: Balancing Precision and Cost in Long Context Retrieval

Late chunking, a novel method introduced by Jina.ai, enhances context retention and retrieval performance. It is now available in Weaviate with minimal code changes!

It offers ColBERT-like functionality but maintains storage costs comparable to standard embedding models.

💡 How late chunking works:

1️⃣ Embed the entire document into multiple token embeddings.

2️⃣ Mark the positions of your chunks (e.g., sentences).

3️⃣ Create late chunk embeddings by pooling token embeddings at chunk positions.

➡️ Result: "Late chunks" retain more contextual information from surrounding text than traditional chunking methods.

Learn more

Stay tuned! Late chunking will soon be a native feature in Weaviate's embedding service.

AI [in Prod] Seattle

One week away! Our in-person roadshow is landing at the AWS offices in Seattle on September 19th.

We know you want to attend the tech talks 💡 But there are a few other reasons you might want to stop by:

🏙️ Show off your Seattle pride with our limited-edition t-shirts

✅ Get a training certificate by completing Zain Hasan’s afternoon workshop

🫶🏻 Make new friends who are just as excited about GenAI as you are

Space is limited — request a ticket here.

AI data tips, tricks, and tech

How-to

Verba: Chat with your data

Edward will guide you through Verba's latest updates, which include:

  • Custom metadata

  • Async ingestion powered by the latest Weaviate client

  • Vector visualization

  • And much more!

Learn how to run free, open-source RAG models and chat with your data in minutes.

Upcoming events

Online

In-person

Remember to check out the provided links for all the details on how to sign up. We can't wait to meet you!

Give us feedback 💚 Q3 Community Survey

We'd love to hear your feedback again!

Our Q3 Weaviate Experience Survey is open until September 30th, 2024. This time, we're eager to learn about:

• How you evaluate models

• Your preferred deployment options

• The use cases you're tackling with Weaviate

Let's build and improve Weaviate together!

Take the survey — to shape Weaviate's future and enhance our community experience.

Welcoming new faces

  • Mary joins us from Germany 🇩🇪 as a Machine Learning Engineer

  • Rodrigo comes from Spain 🇪🇸 and joins us as a QA Engineer

  • Luke is joining us from Georgia, US 🇺🇸 as a Commercial Account Executive

  • Michael comes from Tennessee, US 🇺🇸 as our Marketing Operations and Lifecycle Manager

  • Danny joins us from Germany 🇩🇪 as a Full Stack Engineer

Weaviate open roles

Weaviate has many exciting opportunities available. Join our exceptional team and be part of a company recognized for its strong culture. We're proud to be included in Will Reed's Top 100 list!

Discover even more roles and opportunities on our career page! ✨

Thank you for reading

Do you have questions about Weaviate, vector databases, documentation, or other topics? We'd love to hear from you! Say hello in our Community Slack or join our Weaviate Forum to engage in community conversations. We're excited to see your participation in the coming weeks!

Weaviate is open source. Come by our GitHub repo, and don't forget to give us a star while you're there ⭐

Until next time,

Femke