Hello there,
Recently a client approach me to build a solution that would help to reply to customer emails. The idea was to utilize the internal knowledge they had from past customer questions as well as general product knowledge specific to his business.
His idea was to use https://www.pinecone.io/ as the vector database storing the embeddings for internal knowledge base. I have recently built a proof of project like that, but using a full on code approach. I have used Supabase Postgress with pg-vector extension and some custom parsing tools to text/markdown sources.
Overall approach above while feasible and fun, involved a lot of code meaning the budget for implementation was also way above what my customer was able to pay.
3 weeks OpenAI releases Assistants API with data retrieval and all of sudden all complexity of dealing with embeddings is just gone.
- Use Airtable to setup FAQ and all relevant internal data store
- Use Make.com to load data from Airtable as a markdown file
- Run Assistants API with enabled data retrieval on question from customer
- Get a very reasonable factual answer to send to customer
I do think that OpenAI might have just killed business for most standalone vector database startups... It is amazing how fast the tech is progressing.
You can check out sample implementation below.
https://youtu.be/FA-51TUQvtg
