How Atomicwork Leverages Llamaindex Loaders to Power Data Sources

Before, there was the enterprise consumerization wave. Now, there is the AI Copilot wave.

More and more IT teams are looking into incorporating an AI Copilot in their information management practices, thanks to ChatGPT. The idea is that employees should be able to easily access and utilize knowledge in the flow of their work in Teams and Slack, from the enterprise software they use. And as with most enterprise software, IT is the silent partner in all software purchases that has to be the pilot that figures out how to integrate the tool into the tech stack and make data flow between products.

We’ve already written about how our AI Assistant, Atom, leverages an ensemble of models to understand and respond in an intelligent, friendly manner to questions. But this post is to dive deeper into Atom’s ability to access the knowledge the employee is looking for.

Data, data everywhere

A McKinsey report found that employees spend approximately 19% of their work week just searching for and gathering information. Giving employees that time back so they can spend it on role-based tasks would automatically boost productivity and reduce the frustration that comes from peering into many rabbit holes.

However, from the perspective of the AI Assistant builder, this is a problem that is easy to write about but difficult to solve for a few different reasons:

• In 500+ employee organizations, every team favours their own knowledge management platform which means that Atomicwork, out-of-the-box has to support popular data management tools
• We also need to support multiple data sources and seamless sync and integrations
• We also need to solve for data compatibility, security and quality without compromising on performance.

For all of these needs, we’ve found the best partner to be LlamaIndex. Atomicwork addresses all of these above challenges using LlamaIndex loaders to handle various data formats and structures, ensuring accurate and reliable data retrieval. This capability not only enhances our data management skills but also reduces the risks of errors and improves decision-making processes.

Sample loaders

Some loaders that we use for RAG over different types of data: The NotionLoader and the PDFLoader.

This means that LlamaIndex can pull in unstructured text, PDFs and Notion documents and index the data within them to produce embeddings and generate text chunks that can be stored in our vector database. This makes it easy for Atomicwork to perform semantic searches and look for specific information in a document. We also perform post processing to make sure that there is no sensitive or personally identifiable information in any of the retrieved chunks.

‍‍
A real-world use case

Translating that into real-world use: Imagine an employee asks Atom, our proprietary AI Assistant, a question.

Our work with LlamaIndex is a part of our continued commitment to modernizing enterprise service management and enhancing the employee experience. Jerry Liu, founder of LlamaIndex, shares this vision as well.

At LlamaIndex, we are committed to providing robust and efficient data integration tools that empower organizations to harness the full potential of their data. LlamaIndex offers 150+ data loaders to popular data sources, from unstructured files to workplace applications, through LlamaHub. Our collaboration with Atomicwork exemplifies how our loaders can seamlessly integrate diverse data sources, ensuring consistency, security, and quality. - Jerry Liu, founder of LlamaIndex

This is something that Aparna Chugh, our Head of Product, has written about before as well: “A high-scalability B2B GenAI app, especially one that caters to critical domains like HR, legal, healthcare, finance, and business systems, should be able easily process large volumes of data quickly and adapt to changing business data. LlamaIndex helps us deliver results in real-time (or almost) to end-users.”

By combining our AI Assistant along with LlamaIndex’s data framework, enterprise IT teams don’t have to choose between efficiency and employee experience; they can have their cake and eat it too.