in assistant api
- i have created an assistant api, with retrieval tool enabled
- have added functionCalling - to get screened items with fields like id
- have attached a file - containing 1000+ data
here whats happening is:
- its taking a lot of time to respond
- i require a min of 10 items to be retrieved, even after mentioning it in assistant instruction, it gives around 5-7
- often times it creates its own items - hallucinates
- based on results, i feel ai is not going through all the 1000+ items before deciding the result (im using node openai package, i extract data from db and create a json object and create a file locally and. then upload the created file to openai with purpose assistants, only .json is getting excepted while creating an assistant with retrieval, other formats like .md .txt if not giving mimetype application/octet-stream error)
in vector embedding
i used ada 002, weaviate - the nearText (semantic) search was giving wrong items.