Case Studies
Connect external data to LLMs, no matter the source.
"Carbon has been an amazing partner for DocsBot. With the wide array of cloud chatbot training sources our users are always clamoring for, leveraging Carbon instead of developing each in house has been a huge time and money-saver for us. And the team has been incredibly helpful the whole way, developing new features, APIs, and UX improvments to better meet the needs of our customers. We are looking forward to our continued partnership and adding support for more cloud connections ongoing."
Aaron Edwards
CEO of Docsbot AI
Challenge
With a customer base exceeding tens of thousands, DocsBots’s platform enables customers to build and manage ChatGPT-powered bots. These bots are trained with company content and documentation, allowing them to provide instant answers to even the most detailed questions customers ask. Their bots can be deployed standalone or added directly to websites. Their impressive customer base includes companies such as Sony and Sentry.
When we first began working with DocsBot, users were limited to training their AI chatbots solely on website content and local file uploads. However, the team at DocsBot realized the value of enabling customers to upload content from 3rd-party data sources and quickly understood the value proposition of Carbon.
Solution
We first worked on enabling DocsBot customers to upload content directly from Notion and Google Drive. We have deep integrations with both Notion and Google Drive and had built-in support for Notion databases along with native Google Drive file formats like Google Docs, Google Sheets, and Google Slides.
DocsBot already had an in-house vector database, so we allowed them to retrieve chunks and embeddings via API to store directly on their end. Since DocsBot uses our Carbon Connect component, we have pre-built callback functions that notify them what 3rd-party data sources a user has connected and what files were uploaded. Then they rely on our webhooks to determine when file contents are processed and ready for them to retrieve. Underneath the hood, we handled the message queues, loading balancing, autoscaling based on demand, file parsing, chunking + embeddings.
The successful launch of Notion and Google Drive prompted DocsBot AI to launch more data sources with Carbon. Within a short timeframe (weeks not months), additional data sources such as Dropbox, SharePoint, Box, OneDrive, and Zendesk were added.
Conclusion
The close collaboration between Carbon and DocsBot is just starting. The immediate priority is on rapidly expanding the number of data sources that DocsBot’s bots can use as training data (for retrieval augmented generation).
