(13_PHUNKOD/Shutterstock)
Generation search (RAG) is now accepted part of the Generatif AI (Genai) workflow and is widely used to feed your own data into AI. While rag works, calls to external tools can add complexity and latency, which led people in Mongodb to work with in-database technology to speed up the thing.
As one of the most popular databases on the planet has developed Mongodb integration to support Langchain and Llamiaindex, two popular tools that developers use to create Genai applications. Developers can also use any external vector database that they want to store vector insertion, indexes and questions.
“There are a number of ways to build rag workflows, says Benjamin Blast, the director of the Mongodb product.” But it is basically just adding friction. As a developer, I am responsible for finding the insertion model, gaining access to it, tracking, measurement – all that is associated with dragging in some new tray component. “
While Mongodb users have options, options are not all the same, says Blast. Whenever you go outside the database, you add to the workflow of friction and latency, says, and larger surface surfaces are also more complicated for tracking and repair when something goes wrong.
“On the overall market we see the tone of confusion and complexity of how to build these systems and how to combine things together,” says Blast. “So we try to simplify dramatically.”
Mongodb wants to simplify things by building more of what Genai developers need to rag directly into their database. The company added vector trade through the functionality of Atlas Vector Search in the fourth quarter of 2023 and at the beginning of this year another big step was made in February when it acquired a company called Voyage AI.
Mongodb says its integration inserting AI in Voyage AI and reraining models will lead to simpler architectures of Genai
Voyage AI has developed a number of models inserting and repetitions designed to speed up search information at Genai workload and improve overall application performance. These models are offered to embrace and consider the most modern.
Voyage AI models AI work hand in hand and convert source data to vector insertion, which are stored in the Mongodb vector storage. Voyage AI has developed a number of inserting models for specific cases of use and even specific domains.
“They have inserting range ranges that have different sizes that allow you to choose how good the results will be,” tents Blast tents will be Bigdatawire In a recent interview. “And then we also allow you to choose to use what is called domain -specific models that are well tuned to specific industry data, so you can have one for code or one for finance or one for the law, so it will be even better results.”
Meanwhile, the AI platforms are continuously optimized to ensure insertion to ensure the highest accident while running, for text and image models. These models increase performance by analyzing vector queries and answers and evaluation, which of them are the best. Then he will re -go questions and answers (ie inserting a preliminary vector) to ensure that the best ones are new.
“This will change the set of results and give you the highest accuracy by providing another 5% to 7% of the accuracy of this result,” says Blast.
The combination of vector trade insertion and models that appear and inserting a cruise help customers to tune the workflows to ensure that their foundation models get the data they need to make good decisions early.
“We can do more clever things to surround integration to improve the accuracy of the results of the past, what models give themselves,” says Blast. “We can make truly selective improvisations for this overall workflow, from the model insertion to a database to the index that our customers would either have a lot of mess and would require a lot of complexity or could not do them.”
Mongodb currently brings the Vector Store and Voyage AI to Atlas Mongodb, its managed database offering in the cloud. Vector search will eventually be available as open source; The company was determined whether the AI Voyage models will also be available as an open source, says Blast. Customers can also use AI Voyage with Langchain and Llamiaindex if they like it.
Mongodb is a notorious database friendly database. Other databases are likely to follow its management in creating these types of specialized models inserting and repetition directly into the database. For the time being, however, New York is happy to lead in this department.
“I think we have taken a love unique approach that gives customers the advantage of integration,” says Blast. “You have to use the same set of drivers and other skills to make it easy to use, but on the back, still a scale of scale, which is one of the real advantages of Mongodb.”
Related items:
Mongodb 8.0 release increases the database performance bar
IBM to buy Datastax for database, capability genai
Mongodb automates resharding, adds time series support
(Tagstotranslate) Model insertion