Innovating with Large Language Models: Smile's LLM Hackathon
- 30 May 2023
Smile has always championed innovation and embraced technology's potential to transform the way we work. Keeping up with this spirit, we recently embarked on a journey to explore and utilize Large Language Models (LLMs). These AI-powered models, capable of understanding and generating human-like text, promise to revolutionize various industries and fields, making them an intriguing frontier for innovation. To nurture this potential, Smile organized a two-day hackathon on May 11th and 12th, focusing on the creative applications of LLMs.
The LLM Hackathon was a decentralized event, where participants could join remotely or at a Smile agency. Our goal was to dig beneath the media hype around LLMs, such as ChatGPT, LLaMA, Midjourney, Dall-E 2 or Gen-2, and discover how these AI models can augment our everyday work. We encouraged our teams to explore privacy-aware applications and decide on the best deployment methods for the models - from public APIs to private instances connected to business intelligence.
Teams of three to five people collaborated over two days, choosing from a range of topics, including creating GitHub Copilot like functionalities for project member onboarding, leveraging LLM abilities for home automation, enhancing developer workflows, and building ChatGPT-like assistants for sales. Our focus was not just on creating technically impressive solutions, but on finding innovative applications and delivering user-friendly experiences.
Participants were also supported by resources from Azure and OpenAI environments provided by Smile. We chose Azure due to its robust data protection safeguards and its licensing for running private instances of OpenAI models.
Very dev oriented profiles
The project's concept involves generating product descriptions by crawling the internet for similar product names. The system then asks GPT to create a summary based on various product descriptions. Finally, the end-user can download a user-friendly CSV file containing the product titles and descriptions.
Here we use the power of GPT to propose a personal shopper assistant to whom we can expose a situation ("I'm going climbing next weekend") and who will translate this into a query for the Gally merchandising engine and allow to suggest products. It doesn't stop there, since it is possible to continue the search in a conversational way in order to refine it.
Readme AI generator
This project makes it possible to scan a code repository (GitLab only for the moment) and generate documentation in order to allow a new collaborator to understand the project more easily by providing explanations on what a particular file does and what it contains, and how it interacts with the rest of the code.
The team has integrated, from the beginning, the financial constraints linked to the use of GPT in order to be able to follow and optimize costs.
Smile Knowledge Bot
This project aims to index knowledge, in the broadest sense, at Smile. Whether it's in a wiki, on a Google Site or in documents on a drive.
One of the first challenges that the team faced was to collect information. Indeed, the choice of Google Sites locked the information and knowledge in spaces that are not reachable by APIs and that we protect behind group authentication.
Once this content is indexed, it is possible to request a bot available in our Google Chat group, which provides answers and references to the documents.
Neo, your sales bot
The announced goal of this solution is to provide an assistant to sales people, which allows them to query the CRM, or other data sources, in natural language (and possibly vocally); as well as to prepare them for customer interviews by allowing them to visualize "trick questions" adapted to the context of the customer they will meet.
As the curtain fell on our LLM Hackathon, we were reminded of the power of technology to empower employees in various capacities and roles. It was evident that while there is immense potential in the current technological landscape, there remains a void of practical tools that our employees can utilize to enhance their efficiency and productivity.
While it may be tempting to cater mainly to developers, the hackathon reinforced the importance of broadening our focus. Employees, in the widest sense, are the backbone of our company. Therefore, the creation of inclusive solutions that benefit all is paramount. For instance, centralizing scattered knowledge into an easily searchable database can revolutionize how we retrieve information, decrease reliance on support functions and managers, and improve autonomy across our global offices.
In parallel, there is great value in enhancing the tools available to our sales teams. Improving access to customer data and streamlining the process of identifying customer needs, finding references, and learning from past experiences can significantly elevate our customer interactions. AI-assisted tools that automate correspondence are just the tip of the iceberg, and we are excited to explore how these technologies can address the challenges outlined above.
For our developers, the hackathon was a testament to the abundance of tools and opportunities that can augment their skills and efficiency. Starting with more immediate solutions like GitHub's Copilot or Amazon Code Whisperer, we foresee a wave of tools that can assist in documentation, testing, or code explanations. These can subsequently be integrated into our continuous integration and continuous delivery (CI/CD) pipelines, thereby elevating our overall development process.
The possibilities offered by LLMs are real and concrete, transforming the way people will work, and we are far from the wishful thinking that NFTs and use cases around the Blockchain may have been.
We are incredibly proud of all the participants for their hard work, dedication, and groundbreaking ideas. Their efforts have pushed the boundaries of what we believed possible and given us a glimpse into the future of AI in business. This event is a testament to Smile's commitment to fostering innovation and staying at the forefront of technological advancements. We look forward to seeing the continuing evolution of these projects and the potential transformations they could bring to our company and beyond.
In conclusion, the hackathon served as a fruitful platform for us to understand our technology needs better and explore ways to bridge the gaps. We remain committed to leveraging AI to empower our employees and look forward to the exciting advancements this commitment will bring.