Skip to content

README.md

Latest commit

 

History

History
48 lines (33 loc) · 1.28 KB

README.md

File metadata and controls

48 lines (33 loc) · 1.28 KB
Raw

Smart Email

Our project made answering ai using rag and solar llm. Thanks to powerful embedding of solar, our powerful rag extremely increased accuracy of our answers. Furthermore, we leveraged self-querying which provides high performance answering with meta-data.

Moreover, we tried deidentification of personal informations in order to prevent them from leaking to llm server.

Environment

python 3.11.2

Dependency

https://github.com/LimePencil/Email-RAG/blob/main/requirements.txt

Setup

1. install requirements

pip install -r requirements.txt

2. Download email

Locate your email json file to data/graph_rag/ directory *Unfortunately, due to the privacy problem, we cannot provide our dataset. Thus, there might exist some discrepencies from our test environments.

3. upload email to mongodb

https://github.com/LimePencil/Email-RAG/blob/main/utils/db_upload.py

4. indexing

Upload embedding to elastic cloud
https://github.com/LimePencil/Email-RAG/blob/main/utils/indexing_deidentification.py

Used Upstage Api's

  • Upstage Document OCR

  • Upstage Layout Analyzer

  • Embedding, Solar embedding-1-Large

  • LLM, Solar-mini-chat

  • Groundedness Check, Solar-1-mini-groundedness-check

Test

$ uvicorn main:app --reload

Contributors