Avatar of David BanysDavid Banys

Train an Expert Chatbot with Chatbase

Chatbase helps companies companies build AI agents that are trained on their data to chat with users and perform tasks. We were lucky enough to talk to Chatbase Founder Yasser Elsaid about how Chatbase got started, where it’s finding success, and where it’s headed next.

Embed a chatbot trained on your own data on your own site with Chatbase

Embed a chatbot trained on your own data on your own site with Chatbase

Railway: What is Chatbase? Can you tell us a bit about how you got started and what problem you solve?

Yasser: Chatbase is a platform that helps businesses build chat-based AI agents. We make it easy to create LLM powered chatbots that can do customer support, sales, lead generation, and more.

I started Chatbase in November of 2022 a few weeks before ChatGPT was launched. I was in my final year of university and wanted to give entrepreneurship a shot before I graduated and needed to look for a job.

I saw solopreneurs like Pieter Levels build cool AI products and I was inspired. I started playing around with the OpenAI API and saw a lot of potential. The idea for Chatbase wasn't super hard to come up with. I didn't do any validation or market research since this space was just being born and it was obvious to me that "ChatGPT for your data" would become a thing and that many people would find it useful.

Over time, Chatbase evolved to become a B2B solution for customer-facing AI chatbots.

Railway: How does your technology work? Can you give us a high level overview of what your application is doing to allow users a custom GPT implementation?

Yasser: Chatbase was one of the first SaaS products to do RAG (Retrieval Augmented Generation), which allows us to “train” these LLMs on custom data like files, documentation, websites, etc …

It works by ingesting and indexing user data into a vector database. When a question is asked, relevant information is retrieved from the database and provided to the LLM to be used to generate the response.

This is the basic implementation of RAG but there is a lot more that goes into it like chunking algorithms, query expansion, re-ranking, handling images, and more.

This is the core technology for building AI chatbots, but we've added many other features to make it even more valuable for businesses. These include integrations with tools like WhatsApp and Slack, advanced analytics that track chat topics and sentiment, a ticketing system to escalate complex queries to a human, and AI actions that allow the chatbot to perform tasks like adding items to a cart or booking meetings on behalf of the user or any action specific to your business.

We trained an AI chatbot on our Railway Docs corpus

We trained an AI chatbot on our Railway Docs corpus

Railway: How much of the product vision have you built so far? Where do you want to take Chatbase next? Are there any features on the roadmap that are especially important or exciting to you?

Yasser: I think right now we’re providing a lot of value for our customer, but we still have a long way to go. Our product roadmap is very ambitious as we aim to dominate the AI agent space. One feature we're particularly excited about is 'AI Actions,' which enables your chatbot to perform almost any task.

Here are some of the actions we're working on:

  • Google Calendar Action: This allows the chatbot to book sales calls
  • Collect User Info Action: Enables the chatbot to collect user data for lead generation
  • Booking.com Action: Lets the chatbot display and make reservations directly through chat

The most exciting feature is Custom Actions, where you can provide an API to the chatbot, describe what it does, and prompt it on when to call it. The chatbot handles the rest.

For example, you could have a Railway chatbot that can start a new project or increase replicas for a specific project, all through chat.

Chatbase provides project analytics in the dashboard

Chatbase provides project analytics in the dashboard

Railway: How has the Chatbase team used Railway? What does Railway enable for your team that otherwise might be difficult?

Yasser: We use Railway for our most compute-heavy tasks, like processing big documents and chunking text. It’s been a game-changer because it handles all the autoscaling for us, so as our workload grows, Railway scales up automatically ...

The fast build times are also a huge plus. We can deploy updates and new features quickly, without any downtime. The ease of deployment means we can push changes fast and try out new things without worrying about the infrastructure. Overall, Railway lets us focus more on building cool stuff instead of managing servers.

Railway: If our readers would like to learn more about Chatbase, how can they get started?

Yasser: They can visit our website at chatbase.co to get started and can follow me on Twitter for product updates and occasional memes :)