-
Type:
Task
-
Resolution: Won't Do
-
Priority:
Unknown
-
None
-
Affects Version/s: None
-
Component/s: AI/ML
-
None
-
Python Drivers
-
None
-
None
-
None
-
None
-
None
-
None
Context
I got a DM from an Intern bernard.tam@mongodb.com who works for rachelle.palmer@mongodb.com.
Essentially, the ask is to create a Python package that performs Vector Search free from all the dependencies of LangChain. These is a very reasonable request. Adding a bunch of new dependencies to a production environment can be a heavy burden, one that I have had to bear numerous times.
IMHO it is testament to the maturity of LLMs that this request is finally coming in. Let's discuss: prakul.agarwal@mongodb.com , steve.silvester@mongodb.com , james.kovacs@mongodb.com. This could be a Q1 task.
Definition of done
Determine scope. At a minimum, produce the basic RAG performed in
MongoDBAtlasVectorSearch. (Add text documents, create indexes, perform HNSW search.) Swinging to the other side, we could provide most of our functionality through this package, and simply wrap it in integrations like langchain and llama-index.
Pitfalls
TBD
Bernard's DM
Hey Casey! My name is Bernard and I'm an intern working under Rachelle in ModFac FE. Nice to meet you!
11:01
bernard.tam
Rachelle informed me that you worked on the implementation for the langchain-mongodb Python package. If you don't mind, I have a few quick questions about that.
11:01
bernard.tam
I'm currently working on a AI tool to be used by ModFac during incubators, and it makes use of langchain-mongodb for its vector search functionality, connecting to an Atlas Vector Search deployment. However, a significant concern our team has with the current setup is the many dependencies langchain itself relies on, which has caused problems in the past with clients in sensitive industries like banking (i.e. Wells Fargo).
11:02
bernard.tam
In light of this, I have come up with a few potential ideas I would like to seek your wisdom on: * in your opinion, is it feasible for us to create a stripped-down/lightweight version of the langchain-mongodb package with minimum dependencies but the same functionality we need (only MongoDB vector search and vector database population) or does the langchain-mongodb package heavily depend on langchain itself such that it cannot be decoupled in such a way?
- is there another package out there you could recommend that has the same functionality we need (only MongoDB vector search and vector database population)?
11:02
bernard.tam
Thank you so much in advance! I look forward to hearing from you soon