Pranav Varshney

Ratna Global Technologies / Engineering case study

Company knowledge for support. Caching for efficiency.

A RAG-based customer support system grounded in company data, plus Redis caching to reduce API costs in a React application.

Team
Client Automation
Role
Software Engineering Intern
When
May – August 2023

The work, at a glance

As a software engineering intern in Client Automation at Ratna Global Technologies, I built a customer support system using company knowledge and added caching to reduce redundant API requests.

I.

Customer support

Retrieval-augmented generation (RAG) using company data to answer customer inquiries.

II.

API efficiency

Redis-backed caching for a React application to reuse responses and reduce request costs.

Customer inquiries resolved
80%
Support-system accuracy
90%
Monthly API cost savings
$1.2K

01 / Customer automation

Ground support responses in company data.

I developed a retrieval-augmented generation (RAG) customer support system using company data. The system resolved 80% of inquiries with 90% accuracy.

The workflow retrieves relevant company context for response generation, connecting the support answer to the knowledge available within the company.

01 / Project workflow

From company knowledge to support answers

Full size
Customer inquiries and company data feed retrieval and generation to produce a support response. The system resolved 80 percent of inquiries with 90 percent accuracy.
Conceptual RAG workflow. Inquiry resolution and accuracy are separate measures of the support system’s results.

02 / Application performance

Reuse responses to reduce API traffic.

I added Redis-backed caching to a React application, reducing redundant API calls and saving $1,200 per month in API request costs.

A cached response can satisfy a repeated request without another API call. The diagram shows the cache-hit and cache-miss paths at the application level.

02 / Project workflow

Cache hits avoid repeated API calls

Full size
A React application uses a Redis cache lookup. Hits reuse stored responses; misses request the API and store the response before returning it to the application.
Conceptual request flow for the caching work. Redis is shown as a separate cache used by the React application; a cache hit avoids a redundant API request.

03 / Impact

Track support quality and recurring cost.

The support system resolved 80% of inquiries with 90% accuracy. The caching project reduced repeated API requests, saving $1.2K each month.