Building intelligence through the silence.

I'm Hritik Kumar — ML Data Ops at Amazon, working across multi-modal data and shipping LLM-powered products. Quiet, precise engineering for systems that have to be right.

Begin Journey

About

A quiet obsession with data quality.

I work at the intersection of machine learning operations and product engineering — labeling and validating multi-modal data at Amazon by day. I care about error trends, process gaps, and the small improvements that compound into accurate models.

Education

Bachelor of Technology, Information Technology

Techno Main Salt Lake, Kolkata · 2020 — 2024 · 7.98/10 CGPA

Experience

Where I've been building.

Go-AI Associate II (ML Data Ops.)

08/2025 — Present

Amazon · Noida

Working with multi-modal data (text, images, speech, and video) to train AI models.
Ensuring high-quality data labeling crucial for Amazon's AI accuracy.
Analyze error trends, identify process gaps, and propose smart improvements.

Software Engineer Fellow

08/2024 — 06/2025

HeadStarter AI · Remote

Built 5+ AI apps and APIs using Next.js, OpenAI, Pinecone and Stripe with 98% accuracy, seen by 1000 users.
Developed projects from design to deployment alongside 4+ engineering fellows using MVC design patterns.
Grew cross-functionally through Agile, CI/CD, Git and microservice architecture.

Software Engineer Intern

11/2023 — 06/2024

Reality AI Lab · Remote

Worked with OpenAI, Vertex AI and TensorFlow to enhance ReX, an AI coach.
Collaborated with the team to implement AI algorithms and improve existing models.

Projects

Systems, not demos.

Hiver AI Email Reply System

An end-to-end system for generating and evaluating AI-powered customer support email replies for Gmail-based shared inboxes.

Built a synthetic dataset of 50 email threads across 10 support categories — billing, technical support, escalations, churn and more.
Developed a reply generator using GPT-5-mini with a specialized prompt for concise, empathetic, professional responses.
Designed a multi-dimensional LLM-as-judge evaluator scoring key-point coverage, tone, accuracy and conciseness.
Achieved an average evaluation score of 79.85/100 across the generated reply set.

PythonOpenAI API (GPT-5-mini)JSON dataset pipeline

View on GitHub

Skills