Gleb Blumental

Key Skills

Software

Other

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Text

Video

Top Label Types

Computer Programming Coding

Evaluation Rating

Fine Tuning

Prompt Response Writing SFT

Question Answering

RLHF

Text Generation

Text Summarization

Translation Localization

Freelancer Overview

I have hands-on experience evaluating and refining AI and LLM-generated responses, including designing multi-turn conversations and real-world user–assistant prompts for advanced language models like GPT-4 and Gemini. My background includes troubleshooting ML outputs, conducting RLHF training for LLMs, and improving the quality of AI-generated data. I am skilled in Python, cloud platforms (AWS), and agile workflows, and have contributed to both backend and frontend development projects. My strong communication skills, combined with a solid understanding of instructional design and user interaction, enable me to deliver high-quality, accurate training data for AI systems. I am passionate about contributing to global Open Science and ensuring the reliability and effectiveness of AI models.

Labeling Experience

Video dub evaluatio

OtherVideoEvaluation Rating

We are evaluating the video dubbing in our assigned language, by rating the quality of it. We have to differentiate the AI dub against the real actors that are dubbing the video and rating these on a scale from 1 to 5, where 1 is either unlistenable dub or an AI generated dub and 5 is a professional actor dub.

We are evaluating the video dubbing in our assigned language, by rating the quality of it. We have to differentiate the AI dub against the real actors that are dubbing the video and rating these on a scale from 1 to 5, where 1 is either unlistenable dub or an AI generated dub and 5 is a professional actor dub.

2025

AI Response evaluator

Scale AITextQuestion AnsweringText Summarization

There are multiple projects I've worked on, which all fall into the same category of RLHF training and evaluation of AI models. I have written prompts and evaluated responses by multiple criteria, mostly languagewise. I have also worked on coding tasks, where I've evaluated the model's coding response and if it has worked, by evaluating it by mostly the same criterias, including codewise criterias.

There are multiple projects I've worked on, which all fall into the same category of RLHF training and evaluation of AI models. I have written prompts and evaluated responses by multiple criteria, mostly languagewise. I have also worked on coding tasks, where I've evaluated the model's coding response and if it has worked, by evaluating it by mostly the same criterias, including codewise criterias.

2024

LLM Trainer

OtherTextText GenerationRLHF

The project was an LLM training project, where we taught the model to use the provided tools, by writing a system prompt, then generating step by step natural conversation, where we had to evaluate the model's response, it's tool usage and it's ability to follow the generated information from the JSON's that were provided in the chat. Then we had to correct and fix the LLM errors, which were both response and tool usagewise. Lastly, on that same project I've worked on evaluating the safety of the same model, such as jailbreaks, prompt injections and unsafe requests.

The project was an LLM training project, where we taught the model to use the provided tools, by writing a system prompt, then generating step by step natural conversation, where we had to evaluate the model's response, it's tool usage and it's ability to follow the generated information from the JSON's that were provided in the chat. Then we had to correct and fix the LLM errors, which were both response and tool usagewise. Lastly, on that same project I've worked on evaluating the safety of the same model, such as jailbreaks, prompt injections and unsafe requests.

2025 - 2025

Education

Work History