RAG-as-a-Service platform Ragie takes flight to bridge corporate data and AI

Retrieval Augmented Generation (RAG) has become an essential part of enterprise AI workflows, but how easy is it to actually implement?

That’s the challenge that startup Ragie is looking to solve. The company is officially launching its eponymous RAG-as-a-service platform today with a generally available release. Ragie is also announcing a $5.5 million seed round of investment led by Craft Ventures, Saga VC, Chapter One and Valor. The promise of Ragie is as a simple-to-implement, yet powerful, managed RAG platform for enterprises. RAG connects enterprise data with generative AI large language models (LLMs) to provide updated and relevant information.

Although Ragie is a new company, its technology is already in use as a core element of the Glue AI chat platform, which launched in May. The founders of Ragie had been working on RAG applications and realized there was a major problem with quickly cobbling together data pipelines.  When Glue reached out to them, it seemed like a good opportunity to start Ragie and solve Glue’s problem.

“We had been experimenting with RAG and realized that there was really just this big blocker when it came to getting data into the system,” Bob Remeika, CEO of Ragie told VentureBeat.

Remeika explained that at its core, Ragie is a data ingest pipeline. It easily allows developers to connect their data sources, which can include common business locations such as Google Drive, Notion and Confluence. The Ragie system ingests those data sources and then optimizes the data for vector retrieval and RAG applications.

Ragie offers a free plan for developers to build and experiment with their AI applications. When developers are ready to deploy their applications to production, Ragie charges a flat rate of $500 per month with some limits. For customers that exceed 3,000 documents, Ragie will discuss enterprise-level pricing.

Moving beyond just a vector database to an enterprise RAG service pipeline

There is no shortage of vendors today that offer different types of technology approaches to support RAG. 

Nearly every major database vendor supports vector data, which is essential for generative AI and RAG. Some vendors like DataStax for example with its RAGstack  and Databricks provide optimized RAG stacks and tools that integrate more than just vector database capabilities. What Ragie is aiming to do is a bit different.

The promise of Ragie according to Remeika is a managed service approach. Where organizations and the developers that work for them, don’t have to put together the different pieces to enable a RAG pipeline. Rather the Ragie service is a turnkey approach where developers simply connect via a programmatic interface to enable a data pipeline for a RAG application.

How Ragie works to simplify enterprise RAG deployments

The Ragie platform integrates multiple elements needed by enterprise applications.

Here’s how Ragie works:

Data Ingestion: Ragie allows companies to connect to various data sources like Google Drive, Notion and Confluence to ingest data into their system.

Data Extraction: The platform goes beyond just extracting text from documents – it also extracts context from images, charts and tables to build a rich understanding of the content.

Chunking and Encoding: Ragie breaks down the ingested data into smaller chunks and encodes them into vectors, which are then stored in a vector database.

Indexing: Ragie builds multiple types of indexes, including chunk indexes, summary indexes and hybrid indexes, to enable efficient and relevant retrieval of the data.

Retrieval and Re-ranking: When a user query comes in, Ragie retrieves the relevant chunks and then uses an LLM-based re-ranking system to further improve the relevance of the results before returning them to the user.

According to Remeika, this multi-layered approach to data ingestion, processing and retrieval is what sets Ragie apart and helps reduce the risk of hallucination in the generated content.

Why semantic chunking, summary indexes and re-ranking matter for enterprise RAG

When it comes to the enterprise use of AI, relevance and accuracy are primary goals. After all, that’s what RAG is all about, bringing the most relevant data together with the power of AI.

To that end, Ragie has placed a particular technical emphasis on innovation on the retrieval portion of the platform.

“We put a lot of effort in making sure that we can retrieve the most relevant chunks for generation and that requires building multiple indexes, summary indexes, hierarchical indexes and re-ranking,” Mohammed Rafiq, co-founder and CTO of Ragie explained to VentureBeat.

One area of innovation that Ragie is exploring is the concept of semantic chunking. Semantic chunking refers to a different approach to breaking down the ingested data into chunks, compared to the more traditional method of using a fixed chunk size with some overlap.

Rafiq explained that Ragie uses multiple types of indexing to improve enterprise RAG relevance. At the first layer are chunk indexes which are created by encoding the chunks of data into vectors and storing them in the vector database. On top of that are summary indexes for every ingested document which is used to increase the relevancy of the retrieved results and ensure the final responses come from a variety of documents, not just one. 

The platform also integrates hybrid indexes. Rafiq explained that the hybrid index allows Ragie to provide both a keyword-based and semantic, vector-based approach to retrieval. He noted that the hybrid index provides flexibility in how Ragie can search and rank the most relevant content.

Overall the key goal is to help enterprise developers build with RAG. 

“What we’re doing is really helping engineers get their AI applications built really fast,” Remeika said.