Challenge
Create a working example of an AI-powered product that made it possible to:
1. Handle the first initial interaction of a user visiting the website, enabling visitors to explore First Line Software’s services by asking the website assistant questions
2. Showcase a live example of First Line Software’s ability to build a cutting-edge, AI-generated product.
Create a working example of an AI-powered product that made it possible to:
1. Handle the first initial interaction of a user visiting the website, enabling visitors to explore First Line Software’s services by asking the website assistant questions
2. Showcase a live example of First Line Software’s ability to build a cutting-edge, AI-generated product.
Solution
We assembled an internal product team composed of a product owner, a business analyst, several developers, and technical leaders within the organization. After explaining the vision of the AI-assistant and its intended capabilities, our technical team went to work exploring various large language models to identify the right technology. In the meantime, our product owner and business analyst documented user stories to define the MVP and future sprints.
The Architecture
The technology stack chosen for the project included ChatGPT 3.5 as the core language model, and Azure as the hosting solution. The backend infrastructure featured an administrative panel, which served as a staging environment for testing new features, a chat history tab for storing visitor interactions, and a documents panel housing the content used to generate intelligent responses.
To train the AI assistant, our team utilized ChatGPT to summarize the content from their website. From there, we validated the content summaries were accurate, edited where necessary, and loaded the content into the admin panel. This comprehensive content loading was crucial to ensure the assistant could provide accurate and context-aware responses.
Before the official launch, we also utilized the company’s quality assurance testing services to fine-tune the AI assistant's responses. This iterative process helped identify areas for improvement, ensuring that the assistant could effectively engage with visitors.
We assembled an internal product team composed of a product owner, a business analyst, several developers, and technical leaders within the organization. After explaining the vision of the AI-assistant and its intended capabilities, our technical team went to work exploring various large language models to identify the right technology. In the meantime, our product owner and business analyst documented user stories to define the MVP and future sprints.
The Architecture
The technology stack chosen for the project included ChatGPT 3.5 as the core language model, and Azure as the hosting solution. The backend infrastructure featured an administrative panel, which served as a staging environment for testing new features, a chat history tab for storing visitor interactions, and a documents panel housing the content used to generate intelligent responses.
To train the AI assistant, our team utilized ChatGPT to summarize the content from their website. From there, we validated the content summaries were accurate, edited where necessary, and loaded the content into the admin panel. This comprehensive content loading was crucial to ensure the assistant could provide accurate and context-aware responses.
Before the official launch, we also utilized the company’s quality assurance testing services to fine-tune the AI assistant's responses. This iterative process helped identify areas for improvement, ensuring that the assistant could effectively engage with visitors.
Results
After less than two months, the First Line Software team was able to go from idea to live product. The assistant successfully met its initial goals, serving as a virtual front desk for visitors and offering a working example of the company’s ability to work with generative AI and large language models. Creatively named “Jaime” to subtly incorporate "AI" into the assistant's identity, the AI-assistant features a professional and approachable persona.
Within the first 60 days post-launch, the AI-powered assistant engaged in over 200 conversations, guiding users to relevant information about the company's services, culture, initiatives, and more.
After less than two months, the First Line Software team was able to go from idea to live product. The assistant successfully met its initial goals, serving as a virtual front desk for visitors and offering a working example of the company’s ability to work with generative AI and large language models. Creatively named “Jaime” to subtly incorporate "AI" into the assistant's identity, the AI-assistant features a professional and approachable persona.
Within the first 60 days post-launch, the AI-powered assistant engaged in over 200 conversations, guiding users to relevant information about the company's services, culture, initiatives, and more.