Skip to content
~/harsh
open to opportunities

Harsh Sandhu

Backend / Systems engineer with applied AI

Bachelor of Engineering (Computer Science & Engineering) — Chitkara University

I build backend systems that hold up at scale, and AI that holds up in production.

View projects
  • Based in India
  • Open to relocation
  • Visa sponsorship welcome
status — live

certaws certified solutions architect*

nowsde @ optmyzr

timeIST · UTC+5:30

0+

yrs experience

Stack

  • c#/.net
  • system design
  • data pipelines
  • graphql
  • lora/qlora
Let's talk

About

I'm a backend and systems engineer with close to four years at a B2B SaaS company, where I own the merchant-feed tooling suite that runs across 18,000+ feeds and 100M+ SKUs. Most of my work is the kind that quietly matters: keeping large-scale data systems correct, fast, and predictable, mainly in C#/.NET and GraphQL with a data layer on AWS. I also build at the applied-AI edge of that, from fine-tuning LLMs with LoRA/QLoRA for production use to a from-scratch 3-bit KV-cache quantisation project. I'd rather ship something correct and maintainable than something clever I can't stand behind. I'm open to relocation and would need visa sponsorship, and I'm happy to talk timelines.

education

  • Bachelor of Engineering (Computer Science & Engineering)

    Chitkara University

    2019 – 2023 · 9.84/10 CGPA

Experience

Where I've shipped, and what changed because of it.

  1. HEAD → current

    SDE · Optmyzr

    Jun 2023 — Present

    Remote · Full-Time

    Joined Optmyzr as an intern and now own the company's full merchant feed tooling suite. Built its most-used ecommerce product, a smart product labelling tool, single-handedly, and serve as the ecommerce team's point person for proof-of-concept work, turning raw ideas and API research into production tools. Rated as exceeding expectations in every quarterly and annual review.

    • Own Optmyzr's full merchant feed tooling suite: audit and analytics products running across 18,000+ merchant feeds, ~90% of the customer base
    • Built the company's most-used ecommerce tool, a smart product labelling system, single-handedly: custom AND/OR rule engine, performance segmentation, input filtering, and an analytical AI layer on DuckDB and GraphQL
    • Scaled feed processing from a single product to 100M+ SKUs via a chunk-based pipeline with flat memory use
    • Turned feed audits from problem detection into one-click resolution, with AI-generated summaries and fix suggestions
    • Delivered a unified Shopping analytics dashboard and a peak sale-day auditing dashboard with the ecommerce team
    • Fine-tuned and deployed custom Hugging Face models for company-specific use cases
    • C#
    • .NET
    • React
    • GraphQL
    • DuckDB
    • MySQL
    • Redis
    • AWS S3
    • Hugging Face
    • LLM fine-tuning
    • REST APIs
    • data pipelines
    • system design

  2. SDE · Optmyzr

    Apr 2022 — May 2023

    Hyderabad · Internship

    Spent 1+ years at Optmyzr as the primary developer behind a feed audit product that went from zero to one of the first of its kind in the market. Worked within the ecommerce team but owned the tool end to end: architecture, implementation, and delivery. Converted to full-time on the back of strong reviews.

    • Built the first version of Optmyzr's feed audit product end to end in C#/.NET, from architecture to release
    • Designed a chunk-based processing pipeline that scaled audits from a single product to 10M+ SKUs
    • Built the disapproval alert system that flags Merchant Center policy violations before they hit live campaigns
    • Shipped one of the first feed audit tools in the market, now a foundation across Optmyzr's retail base
    • C#/.NET
    • React
    • S3
    • MySQL
    • Redis

Skills

Languages

  • C#
  • JavaScript
  • Java
  • Python

Backend

  • ASP.NET
  • Node.js
  • REST
  • GraphQL

AI / LLM

  • LoRA
  • QLoRA

Cloud & DevOps

  • AWS
  • Docker

Databases & Cache

  • MySQL
  • MongoDB
  • Redis
  • DuckDB

Frontend

  • React.js
  • Next.js

Systems

  • System design
  • Data pipelines
  • Large-scale processing

Selected work

Things I've designed, built and shipped — click any card for the full story.

3-Bit KV Cache Quantization for LLMs (TurboQuant-Inspired) screenshot
P-01Python · PyTorch

3-Bit KV Cache Quantization for LLMs (TurboQuant-Inspired)

A from-scratch PyTorch benchmark of TurboQuant-inspired 3-bit KV cache quantization, compressing the cache ~4.9× while holding key reconstruction above 0.999 cosine similarity. Measured across the Qwen2.5 family on consumer hardware.

  • PyTorch
  • Python
  • LLM
View project
Loops vs. Prompts: When Iterating an LLM Is Worth the Cost screenshot
P-02Python

Loops vs. Prompts: When Iterating an LLM Is Worth the Cost

A controlled experiment on when it pays to run a language model in a loop instead of asking once. Four strategies, three tasks, matched compute, and proper statistics, all reproducible from the code and data.

  • LLMs
  • Ollama
  • Qwen2.5
View project

Writing

Follow on Medium

Occasional deep-dives and lessons learned from things I've built.

» 20 Jun 2026

I Simulated 100 Indians Debating AI and Jobs for 20 Rounds

They never reached consensus. But halfway through, one thing flipped their conclusions almost instantly, and it wasn’t new data. A few weeks ago I got tired of a question I keep seeing argued badly: will AI replace human jobs in India by...

Read on Medium

Let's talk

Hiring, collaborating, or just curious about something I built — my inbox is open.

harshsandhu913@gmail.com

usually replies within a day

Find me elsewhere

Your details are used solely to respond to you — never shared, sold, or added to a mailing list. (GDPR-friendly.)