Skip to content

NGNischay GuptaSolution Designer

Work About Résumé

Read this site asAIJavaArch

Let’s talkLet’s talk

01Home 02Work 03About 04Résumé 05Contact

Read this site as

Read this site asAIJavaArch

Contact · Open to architecture, AI platform and backend engagements

Bring me the messy version.

Half-formed ideas, systems that buckle under load, AI features that need to be cheaper or safer. That's my favourite starting point.

or open your mail app →

Based in: Surat, India
Local time: IST (UTC+5:30)
Elsewhere: LinkedIn ↗GitHub ↗

01 · What are we building?

Architecture reviewAI / RAG platformBackend & event-driven systemsSecurity & SSOLead a buildSomething else

02 · Timing

ExploringThis quarterSoon as possible

03 · About you

Your nameEmail for my replyThe project, in a few linesCompany

Goes straight to my inbox. No newsletter, ever.

Home Work About Résumé LinkedIn GitHub

© Nischay Gupta · Set in Fraunces & Geist · Built with Next.js · No cookies, no trackers · Press ? for shortcuts

Nischay Gupta

All work05 / 07

AI Audio Analytics & Facial Recognition

Millions of support calls scored by LLM, and a face matcher that searches 10,000 images in under 10 seconds.

Role: Software Engineer
When: 2024
Where: Xtreme Thoughts SoftTech
Client: Xtreme Thoughts SoftTech

agreement with human reviewers
face recognition accuracy
to match 5,000–10,000 images
call-analysis throughput

01The system, drawn
02Audio analytics
03Facial recognition
04Stack

The system, drawn

The system, drawn

Fig. 03A call, split by speaker and scored · a face, matched across the gallerySimplified redrawing · illustrative data · real mechanism

How it works · 2 parts

01
Audio analytics
C#/.NET platform analysing millions of support calls at ~2x previous throughput, with an LLM evaluation layer on OpenAI APIs.
02
Facial recognition
Optimised OpenCV + CUDA embedding pipeline.

Stack

C#
.NET Core
Python
OpenAI APIs
OpenCV
CUDA
SQL Server

Go deeper

Want the full walkthrough?

I’ll take you through how Audio Analytics & Face Recognition is put together: the architecture, the trade-offs and the parts that were hard, then sketch how the same approach would fit your system. Thirty minutes, a whiteboard, no slides.

Talk through a solution Replies within two working days · IST

Next projectUser Management SystemCentralised identity and access for a marketplace, serving millions of API requests.Keep reading