Aman Kota

Key Skills

Software

Labelbox

Other

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Document

Text

Top Task Types

Audio Recording

Bounding Box

Classification

Evaluation/Rating

Prompt + Response Writing (SFT)

Freelancer Overview

I’ve been part of a Quality Assurance (QA) program where I work on priority projects and enjoy additional perks compared to regular contributors in a particular platform. In this role, I’ve dedicated myself to refining and evaluating AI models across multiple domains, ensuring every output meets the highest standards of quality and alignment. Over time, I’ve grown into the roles of Reviewer and Senior Reviewer on several projects, taking ownership of auditing tasks, maintaining quality control, and making sure only the most accurate and reliable data passes through. My work spans evaluation, prefernce ranking, prompt generation, rubrics creation and SFT-driven projects for large language models (LLMs), where I design and apply structured industrial rubrics to assess model accuracy, coherence, and adherence to instructions. I've also contributed to Reinforcement Learning from Human Feedback (RLHF) workflows by evaluating model outputs, identifying edge cases, and creating golden responses that serve as benchmarks for ideal performance. In addition, I've developed and tested STEM and non-STEM prompts designed to push models to their limits and provide fail cases of models in client-specific requirements. Beyond text-based work, I've also handled precision verbatim audio transcription projects, achieving 1-5 millisecond accuracy for 2-minute clips using Labelbox, while consistently maintaining top-tier quality ratings.

Labeling Experience

Audio annotation

LabelboxAudioPoint Key PointSegmentation

The project involves labeling background noise, assistant speech, and user speech. Each word (token) must be annotated with a temporal precision of 1–5 milliseconds. In addition to labeling, we also need to perform judgment tasks - identifying whether the user’s speech occurs as an interruption, standard response, or acknowledgment, regardless of whether we transcribe the speech verbatim.

The project involves labeling background noise, assistant speech, and user speech. Each word (token) must be annotated with a temporal precision of 1–5 milliseconds. In addition to labeling, we also need to perform judgment tasks - identifying whether the user’s speech occurs as an interruption, standard response, or acknowledgment, regardless of whether we transcribe the speech verbatim.

2025

Education

Work History