How Litmos is using Llama to make corporate learning more intuitive

Litmos, a corporate learning platform, is transforming the way organizations upskill their workforces worldwide, with 4,000 companies trusting its innovative approach. The company’s learning management system (LMS) and content libraries enable organizations to deliver learning content to 30 million people in 35 languages. With a vast library of multimedia resources, including video, audio, PDFs, and Sharable Content Object Reference Model (SCORM) files, Litmos streamlines the creation, curation, and connection of content for learning across diverse formats.

While SCORM files are commonly used in edtech due to their portability across platforms, learners sometimes had difficulty discovering content in Litmos’ LMS. Traditional keyword-based search functions made navigation difficult, requiring users to spend excessive time finding relevant courses, which resulted in decreased engagement and lower LMS adoption rates.

Built with Llama 3.1 8B, Litmos created an AI-powered assistant that enables natural language search, making the LMS more intuitive and user-friendly. Now, learners can easily access training, validate their learning, and ask questions that help them perform their jobs more confidently and efficiently.

“Keyword searches are not how humans naturally communicate, and employees don’t spend as much time in an LMS as the administrators, so navigating the system takes them more time than necessary to locate courses or an answer to an open-ended question,” says Tommy Richardson, Chief Product and Technology Officer at Litmos.

Because the Litmos LMS is fundamentally content-driven—with language at its core—the team recognized that large language models (LLMs) were the natural path forward in making learning more intuitive and engaging.

The Litmos AI assistant was designed to encourage learners to explore topics of interest, ask questions, and seek out additional upskilling opportunities to enhance their learning experience. It also generates self-study course summaries and follow-on learning materials for the training cycle, making reviewing and applying critical information easier.

“With the human-like interface introduced through our AI assistant, learners can find the information they need faster and have proven to adopt the LMS tool more routinely,” adds Richardson.

Litmos initially explored a competing LLM to build its AI assistant but quickly pivoted to Llama 3.1 8B when it was released after conducting a comparative analysis. “Llama was a clear choice for us because it delivered better performance, faster response times, and greater cost-effectiveness than competing models,” says Dave Newman, VP of Technology at Litmos.

Litmos’ core use case is generative AI with retrieval-augmented generation (RAG) to power its AI assistant. To ensure a smooth and effective integration, the team first focused on delivering a proof of concept that showcased how AI could enhance LMS functionality. The rollout prioritized a subset of 100 courses and introduced two key features: “assign courses” and “see my courses that are due.” This phase allowed Litmos to validate Llama’s ability to enhance key LMS functionalities before rolling out the final product for content discoverability and learner engagement.

Beyond performance, Llama 3.1 outperformed both a competitor and Llama 2 in instruction following and was more reliable for AI assistant functionalities. The decision to switch was straightforward—Llama 3.1 provided more accuracy and better integration into Litmos’ existing AI infrastructure, without expending additional resources.

The integration enabled more intuitive search, personalized recommendations, and interactive learning experiences within the LMS. By owning and controlling its RAG system, Litmos ensured data protection and privacy for its customers—a critical requirement that only a self-hosted model could provide. This approach alleviated potential client concerns regarding privacy while strengthening its AI capabilities as a long-term asset.

“As an LMS that serves a variety of use cases, including enterprise training, we needed an AI model that could scale while maintaining strict security and privacy requirements. Llama checked all those boxes,” says Tristan Schleining, VP of Security at Litmos.

To bring their vision to life, Litmos partnered with Tribe AI, a firm specializing in building and deploying AI solutions for organizations. Tribe worked closely with Litmos to design, develop, and deploy the AI assistant.

“The Tribe team brought invaluable expertise in AI infrastructure and model integration,” Richardson says. “Tribe’s support has helped us build robust AI products and reposition Litmos as an AI company.”

On Tribe’s recommendation, the AI assistant can both answer and pose questions for learners based on LMS content, reinforcing knowledge retention. For example, a learner can use natural language to ask, “I’m looking for courses that will help me with ‘X’” or “Can you remind me of the key functionality of ‘Y’?” The AI assistant responds intelligently, providing relevant and context-aware answers that drive content discoverability for learners.

Since integrating its AI assistant, Litmos has significantly improved content discoverability and engagement. Llama has provided consistent responses that feel more human than other LLMs with the same class of GPU hardware that were tested.

Thanks to Llama, Litmos expects to cut the time needed to create content by 95% without much loss in quality. They also anticipate a 110% improvement in following instructions, a 59% boost in understanding complex tasks, a 140% increase in professional-level understanding, and a 27% increase in speed.

“More and more, AI in education and corporate training has become table stakes,” says Richardson. “By leveraging Llama, our AI assistant is making learning more intuitive and effective, demonstrating the potential of AI in edtech and corporate training.”