Bright, a trailblazer in the workforce development sphere, offers an innovative approach to corporate learning through immersive experiences, simulation-based training, and AI-aided coaching. Despite their groundbreaking approach, a challenge lay in their path: a protracted client onboarding process stemming from a heuristics-based curriculum development. This approach can often be more time-consuming because it relies on trial and error, intuition, and accumulated knowledge rather than streamlined, standardized processes.
The Bright team had a hypothesis that LLMs and generative AI could help streamline and enhance the training experience, but didn’t know what steps they needed to implement this new technology.
“We knew what direction we wanted to go,” said Tyler Horan, who leads product at Bright, “just not how to get there.”
To address this bottleneck, Bright partnered with Tribe AI.
Bright's strength is in developing realistic training scenarios that mimic authentic workplace situations. However, building out an immersive learning experience including customer situations is an extremely manual process, which limits the speed at which clients can be successfully onboarded.
With the potential of large language models (LLM) in mind, Bright aimed to craft a high quality and customizable training experience that could be up and running in a fraction of the time.
“People learn better when they’re immersed in an experience,” said Horan, “we know that already. But we wanted to use AI to not only mimic what’s already happening and speed it up, but to recreate the experience based on what’s possible thanks to Generative AI.”
“Because Generative AI was so new, there wasn’t a lot of domain knowledge on the open market. It just made sense to work with Tribe.”
“Because Generative AI was so new, there wasn’t a lot of domain knowledge on the open market,” said Horan. “It just made sense to work with Tribe. We needed experts in language modeling who not only knew how existing tech works, but how to learn about what was possible.”
Tribe AI, leveraging its roster of experienced AI talent, brought two experts onto the project:
- Alex W. – NLP and Software Engineer. Ex-Instagram, pioneered text-based clothing recommendation at Stitch Fix, led infra at Megagon Labs, led ML at Academia.edu
- David C. – NLP Researcher with call center expertise. Dialogue systems lead automating human conversations at Gridspace; Ex-Microsoft; MS from Carnegie Mellon
“We started with a lot of ideation and discussion of what was possible,” said David, the Tribe AI researcher on the project. “We prototyped and crystalized scope throughout the project as Bright came to a clearer idea of what made sense for their budget and timeline.”
The Tribe team’s solution was an AI-driven chatbot capable of adopting various customer personas to enrich training scenarios. They scoped and built the following three products, using foundational embedding models, generative models, and retrieval models:
- Dynamically generated customer simulations
- A dynamic, real-time CSR evaluation assistant
- A passive, real-time CSR knowledge assistant
“Our key technical objectives were to build compelling prototype systems for those three aforementioned systems,” said David, “as well as initial exploration of custom modeling from customer documents and data, and an API to build prototypes on top of our models.”
Bright, keen on harnessing the capabilities of large language models (LLM), aspired to create a unique training experience tailored to specific client data. With Alex diving deep into call logs and understanding the nuances of customer interactions, the chatbot was developed to mirror these interactions effectively. Under Alex's guidance, and with David's engineering oversight, the chatbot became a reliable tool in reflecting customer scenarios.
Building on this foundation, Bright added a voice simulation layer to the AI model, further enhancing the interactive experience.
“With the customers who have deployed the product, every single one has said it increased ROI and decreased training costs.”
“Bright’s paradigm is about changing the whole structure of learning in an org and this product really rounds out our ecosystem,” Tyler said. “With the customers who have deployed the product, every single one has said it increased ROI and decreased training costs.”
For Bright, the AI solution is not just an internal tool but a potent instrument for fundraising and sales. Although still in the early stages of deployment, the AI solution has already become an integral part of Bright's demo presentations.
“People have a ‘wow’ moment when they see what it can do,” Horan says. “They say, ‘I didn’t know that was possible.’”
Bright's experience with Tribe AI has underscored the value of collaboration, particularly in accessing the expertise needed for timely market entry. Tribe’s domain expertise accelerated Bright's market entry, bolstering their competitive edge.
“There’s still a lot of space to improve and explore,” said David . “We’ve already planned work with Bright to improve latency and to give shorter, more human-like responses. I think this type of technology is a perfect use case for Generative AI.”
Bright's future vision involves a balance between AI and human interaction in learning experiences. And will they work with Tribe AI again?
“Any time you’re operating on the edge of what’s new and innovative, you probably don’t have the expertise in house,” said Tyler. “What gave us the advantage here was being first to market, so we’ll continue to use expertise like Tribe since it gives us a competitive edge.