Using GPTs as Team Members
Custom GPTs by OpenAI are practically free new hires. But as as new hires go, most are not very useful.
In this post, I share the system we built to turn these new hires into a whole range of experts and how we’re leveraging it in our team to increase the man(AI?)power we have.
Before we dive into the specifics, let me share a few examples to inspire some potential uses:
- All-knowing Customer Support Agent
- On-Call Engineer
- Calls Transcript-Summarizer
- Customer Email Writer
- HR Specialist
Also if you’re already dying to get started, you can skip on directly to our docs and code repository:
Docs: https://docs.danswer.dev/quickstart
Code: https://github.com/danswer-ai/danswer
Let’s start by understanding where GPTs fall short
The first and most obvious is that GPTs are trained on massive web data but they know nothing about your specific team. It doesn’t have the context of your customer interactions, legacy system considerations, engineering designs, and everything else that is crucial to make an informed decision.
The second problem is that GPTs are limited in the amount of knowledge they can ingest. You can’t just pass it all of your documents and hope for the best. This is why Retrieval Augmented Generation (RAG) has been so popular. But even amongst RAG pipelines, there are a lot of variations. For example, the retrieval must respect the user’s access permissions. It must be able to be kept up to date as new information comes in and able to identify out of date sources of information. And there are many important optimizations (for example: discerning between a “relevant” and a “useful” document).
Our solution
We believe that our solution (Danswer), brings value to teams of all sizes so we released it for everyone to use. It is fully opensource (MIT) and self-hostable and you can plug it into any LLM of your choice. It can be used via both a web app and Slack.
The first step in enriching the GPTs is to connect it to your data:
Now your data is connected and indexed! At this point, you can already ask the default AI any question in the UI and it will answer based on your documents.
Now let’s create some custom GPTs! Maybe we want to create an AI on-call agent to help take the load of off your sleep deprived engineers. Let’s tell this GPT what its job is and configure the right set of documents to use for the job.
You can also configure Danswer as a Slack bot. We’re actually using this for our opensource community (we would love to see you there by the way, here’s a join link).
You can find the detailed instructions here: https://docs.danswer.dev/slack_bot_setup
All Resources
Open Source Code: https://github.com/danswer-ai/danswer
Documentation: https://docs.danswer.dev/
Danswer Cloud (Currently in Alpha): https://www.danswer.ai/
Thank you for reading! We’d love to see you in our Slack and Discord! Tell your friends!
