[DMP 2024]: Enhancing Caching and Retrieval for Sunbird-Virtual Assistant

[SumitMakariye]

Ticket Contents

Description

This feature aims to improve the Sunbird-Virtual Assistant by implementing a caching layer for GPT responses and associated documents. By caching responses and leveraging similarity search, the assistant can provide more personalized and efficient responses to user queries.

Goals & Mid-Point Milestone

Goals

• [Goal 1]Implement caching layer for GPT responses and documents.
• [Goal 2]Move system cache into a vector DB.
• [Goal 3]Implement retrieval from cached documents using similarity search.
• [Goal 4]Dockerize the service.
• [Goals Achieved By Mid-point Milestone]

Setup/Installation

No specific setup or installation guide provided.

Expected Outcome

The final product should include a virtual assistant with enhanced performance and personalized responses. Cached GPT responses and documents should be utilized to improve response quality and efficiency.

Acceptance Criteria

GPT responses and associated documents are successfully cached.
System cache is moved into a vector DB.
Retrieval from cached documents using similarity search is implemented.
The service is Dockerized for containerization and deployment.

Implementation Details

Redis for building a caching layer.
Marqo DB as the primary vector DB for storing document vectors.
LLM-based vector similarity search for retrieval from cached documents.
Docker for containerization and deployment.

Mockups/Wireframes

No mockups or wireframes provided.

Product Name

Sunbird-Virtual Assistant

Organisation Name

Bandhu

Domain

⁠Healthcare

Tech Skills Needed

Python

Mentor(s)

Vineet Upadhayay

Category

Backend