genai-bot-framework: insights from Younited on Generative AI

I’m thrilled to offer you a glimpse into our latest foray into generative AI at Younited. This sneak peek is just the beginning, as we prepare to unveil our comprehensive strategies and innovations in the near future. As we chart this thrilling course, I hope to spark your curiosity and excitement for the transformative journey ahead. 😊

𝐀𝐈 𝐚𝐭 𝐭𝐡𝐞 𝐂𝐨𝐫𝐞 𝐨𝐟 𝐘𝐨𝐮𝐧𝐢𝐭𝐞𝐝’𝐬 𝐈𝐧𝐧𝐨𝐯𝐚𝐭𝐢𝐨𝐧

AI is not just a component of our strategy; it’s the foundation of our product innovation, deeply ingrained in our scoring, fraud prevention, KYC processes and various other usecases.

With the surge of generative AI in 2023, we’ve been quick to investigate how these technologies could revolutionize our habits, our Tech and more globally our industry. Our business is technically demanding, with a fully dematerialized, automated, and transparent approach to loan products. Yet, it’s also enriched by the diverse interactions with users and in managing our back-office and customer relations.

𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐢𝐜 𝐈𝐧𝐯𝐞𝐬𝐭𝐦𝐞𝐧𝐭 𝐢𝐧 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐈

Our exploration into generative AI began with assessing how it could be integrated into our Tech Operations to add real value. Despite the higher costs associated with generative AI compared to average API calls, we focused on the potential return on investment. The value derived from these interactions could be substantial, especially in areas where we spend significant time producing information end-to-end: meeting summaries, technical diagnostics, incident communications, support, project management…. Generative AI is ubiquitous, and when not implemented internally, it’s often used indirectly by staff at all levels to not just do the work but to drastically accelerate the production of media, its content, and structure.

We recognized the importance of strategically structuring and providing a technical framework internally to foster and guide the adoption of these emerging generative AI applications.

Younited’s GenAI Evolution and Initial Results

As we embarked on our initial Tech Operation use-cases, it became evident that a one-size-fits-all solution would not suffice for the breadth of applications we envisioned for generative AI. We encountered a landscape of diverse needs and potential integrations, which led us to a pivotal realization: to truly harness the power of AI, we needed a framework that was not just functional but supremely reusable and adaptable. Here’s why we decided to build our own GenAI framework:

1. Diverse Communication Methods: Our operations demanded a framework that could seamlessly integrate with a multitude of communication platforms, from instant messaging and corporate messaging to emails, websites, ticketing tools, and even pure API calls for reuse by other programs. This variety required a flexible backbone that could cater to all these channels efficiently.
2. Complex Task Management: We recognized the need for a system that could handle rich and complex tasks, simulating “real-life” interactions with a high degree of sophistication. To achieve this, we had to design a workflow that could not only accommodate but also enhance the capabilities of AI in managing intricate processes.
3. Customization and Scalability: The existing frameworks, while promising, fell short of our specific requirements. They often necessitated extensive recoding to align with our vision. We needed a framework that could scale with our ambitions and evolve with the rapidly advancing AI landscape without constant overhauls.

Our partnership with Microsoft and the support from Microsoft France were instrumental in this phase, granting us immediate access to the latest OpenAI models on Azure. This access enabled us to dive deep into the capabilities of GPT-4, Turbo, and Vision, among others.

With the commitment from Younited TECH (especially my CTO Stéphane Alizon 🙌) , we have begun to rigorously test our GenAI framework across a spectrum of use-cases. The initial results have been very promising, reinforcing our belief in the necessity and potential of a dedicated AI framework.

Our GenAI-bot-Framework Project

Our genai-bot-framework project, through the use of plugins, covers the three main areas of interaction surrounding generative AI, with two additional critical dimensions that cater to advanced usage and continuous improvement:

1. User Interactions: This includes a variety of communication channels such as Teams, Slack, emails, Jira, APIs, web interfaces, etc., and focuses on understanding and responding to user behaviors.
2. The GenAI Models: We integrate a range of generative AI models like GPT, Mistral, Llama, DALL-E, Stable Diffusion, etc., to provide a versatile set of capabilities for different use cases.
3. Custom Actions: Our framework facilitates custom actions that create a bridge between the prompts of a Large Language Model (LLM) and various activities, such as querying logs, reading internal wikis, interacting with vector databases, and navigating the web.
4. Advanced Usage Development: We’re enhancing our framework to support advanced use cases that allow for seamless workflow integration of different AI models. These can be triggered by human instructions or autonomously directed by the AI itself, enabling a more dynamic and intelligent system that can adapt to complex requirements.
5. Bot Interactions and Human Feedback Management: A pivotal feature of our framework is the ability for bots to interact with each other, creating a more cohesive and collaborative AI environment. Additionally, we’ve embedded native support for human feedback management, which is essential for a structured approach to continuous improvement of both models and bots. This feedback loop ensures that our AI systems evolve and adapt over time, guided by user input and performance metrics.

Sharing Our Journey in Generative AI

We believe in the power of sharing knowledge, especially in the nascent stages of emerging technologies like generative AI. We’re excited about the idea of making our GenAI framework accessible to the community in open source once it’s ready. By doing so, we hope to contribute to the collective development of these technologies.

We’re excited about the potential of generative AI and eager to share our progress. Keep an eye on the Younited blog for more details about our GenAI-bot-framework coming soon 💡.