Ryan Morov

Key Skills

Software

Data Annotation Tech

Labelbox

Mindrift

OneForma

OpenCV AI Kit (OAK)

Remotasks

Surge AI

Toloka

Telus

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Image

Text

Top Task Types

Classification

Computer Programming Coding

Evaluation Rating

Fine Tuning

Freelancer Overview

I'm an experienced AI training contributor with a strong focus on evaluating the programming prompts and code generation tasks. I specialize in labeling computer code, natural language and image-based data, with expertise in classification, rating & evaluation, fine-tuning prep, and debugging logic-based outputs. I've worked with both structured and unstructured data and am particularly skilled at reviewing LLM outputs for syntax correctness, logical coherence, relevance, and general functionality in coding contexts. In addition to prompt evaluation, I have hands-on experience with Python-based data visualization, Unity & C# game development and debugging code from image-based challenges. My background in software development allows me to spot edge cases and subtle technical issues in model behavior. I'm deeply interested in improving model alignment, especially for programming and conversational AI models, and I bring a precise, feedback-driven mindset to every task I take on.

Labeling Experience

Negative constraints in programming prompts

Data Annotation TechComputer Code ProgrammingComputer Programming Coding

The goal is to create a prompt that will get the AI model to fail, using a negative constraint. For example, the prompt tells the AI model not to do something, and yet the model does said thing. The worker must continuously try and modify their prompt until the AI model fails, at which point the worker will grade the response, as well as provide a perfect response to their prompt.

The goal is to create a prompt that will get the AI model to fail, using a negative constraint. For example, the prompt tells the AI model not to do something, and yet the model does said thing. The worker must continuously try and modify their prompt until the AI model fails, at which point the worker will grade the response, as well as provide a perfect response to their prompt.

2024

Fine-tuned criteria in programming

Data Annotation TechComputer Code ProgrammingComputer Programming Coding

This project has a goal of getting the AI models to be able to evaluate themselves. We look at an AI conversation (or create a conversation ourselves), and create a list of fine-tuned criteria, which contains things that must be respected for the response to be considered perfect. This is a very time-consuming project.

This project has a goal of getting the AI models to be able to evaluate themselves. We look at an AI conversation (or create a conversation ourselves), and create a list of fine-tuned criteria, which contains things that must be respected for the response to be considered perfect. This is a very time-consuming project.

2024

Coding prompt generation and response

Data Annotation TechComputer Code ProgrammingComputer Programming Coding

Generate a coding-related prompt (specific rules), wait for the model to respond, check the model's response for correctness, truthfulness, harmlessness, completeness, verbosity, etc. If the model's response isn't perfect, modify the response.

Generate a coding-related prompt (specific rules), wait for the model to respond, check the model's response for correctness, truthfulness, harmlessness, completeness, verbosity, etc. If the model's response isn't perfect, modify the response.

2024

Education

Work History