I am working on creating a chatbot for domain specific application. I want to use GPT 3.5 API exclusively for that. Data is getting added/updated/deleted on daily basis. I am trying to figure out what should be the correct approach to integrate model in my product. Do I need to fine tune model or use embedding for that?
I guess there are different approaches. I have created a chatbot that also interacts with data that is updated on a 24/7 basis and housed in a SQL database. In the case of my bot, I leverage function calling to interact with the database. Depending on the nature of the user question either a semantic search or a SQL query is triggered to retrieve data from the database, which then serves as an input/context for generating a response to the user. As the database is updated automatically, I don’t have to worry anymore about data being up to date.
I’ve used OpenAI’s Assistant to implement this approach.
2 Likes
Hi and welcome to the community!
Fine-Tuning is very likely is not the right tool for the job.
Embeddings may be the right choice but in general this advice applies to text or images.
The suggestion from @jr.2509 sounds actually good but if it’s applicable to your case depends.
In order to give you good advice it would be helpful to know what type of data we are talking about. What format is the data in? Are there several types of data? How does the update process look like? If you describe your use case in a bit more depth this could help.
1 Like
Thanks for the reply. I am using sql and I have only text and numbers as part of data types. Data is getting updated on daily basis directly in the database. Thanks.
Thanks. I will also try assistant approach and update here. Thank you again.
1 Like