Case Studies
Connect external data to LLMs, no matter the source.
"When we initially launched SiteGPT, we only had the option of training the chatbot using website links. But our customers wanted something more. They wanted to also upload their files to SiteGPT along with their websites. That’s when we came across Carbon. Carbon helped us bring this functionality to life super quickly. All the complexity around reading, and managing large files, handling different file types is being handled by Carbon seamlessly. Because of this, we were able to implement file upload functionality within just a few days. Looking forward to making use of Carbon even more and free up our time to build other things that will bring even more value to our customers."
Bhanu Teja
CEO, SiteGPT
Challenge
SiteGPT has tens of thousands of customers who utilize their platform to build AI chatbots that are experts on the products they’re selling. These AI chatbots are primarily used as user-facing customer support assistants on company websites and messaging platforms such as WhatsApp and Facebook Messenger. SiteGPT also offers multiple advanced functionalities such as lead generation, optional human escalation, and in-app actions.
When we started working with SiteGPT, customers could only train their AI chatbot on their website’s content. File upload was a feature highly requested by customers, but it was more difficult to implement given the infrastructure complexity.
Solution
Carbon helped SiteGPT quickly enable file uploads, including support for PDFs, CSVs, and TXT files. We solved SiteGPT's challenge by providing a pre-built File Upload component in React that users can use to upload files. Additionally, we exposed an API endpoint for retrieving relevant file chunks based on a search query. Underneath the hood, we handled the message queues, loading balancing, autoscaling based on demand, file parsing, chunking + embeddings.
The launch of file uploads was successful, and SiteGPT expressed interest in adding our pre-built data connectors next. They also wanted to leverage our white-label service and use their custom branding across all user-facing components. Within a couple of weeks, we launched the Google Drive and Dropbox connectors, allowing SiteGPT customers to train their AI chatbots using content from these data sources. Carbon supported popular file formats and data source-specific formats such as Google Docs, Sheets, and Slides. With Google Drive and Dropbox connectors being well-received by SiteGPT customers, we subsequently helped SiteGPT launch OneDrive, SharePoint, Zendesk, Box, and Notion.
Conclusion
We have established a great working relationship with SiteGPT and are continually finding new opportunities to expand our collaboration. Our current focus is on having Carbon provide more web scraping infrastructure for SiteGPT. Additionally, we intend to integrate our hybrid search with SiteGPT's AI chatbots, allowing them to more accurately search through all files and data sources to find relevant content.
