Liverson Mwasho

Key Skills

Software

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Text

Top Label Types

Classification

Question Answering

Text Generation

Data Collection

Freelancer Overview

I am a machine learning expert with over five years of experience specializing in generative AI, computational problem design, and AI training data creation. My background includes designing complex STEM problems and Python-based tasks that require advanced reasoning and computational skills, ensuring high-quality data annotation and verification using tools like NumPy, Pandas, SciPy, and scikit-learn. I have contributed to AI model training, refinement, and evaluation in both research and production environments, focusing on improving model reasoning and response accuracy. I am adept at documenting clear problem statements and solution logic for AI datasets, and I thrive in fast-paced, remote settings where attention to detail and adaptability are essential.

Labeling Experience

Generative AI Response Evaluation and Reasoning Annotation

Scale AITextClassificationQuestion Answering

Worked on Generative AI data labeling project oriented towards assessing, grading, and refining AI text responses. These tasks included checking the outputs of the models in terms of facts, logical connections, depth of reasoning and compliance with the detailed instructions. Conducted RLHF-style assessments, where several model responses are compared to the same prompt and ranked in terms of quality, accuracy and clarity. Determined hallucinations, poor reasoning chains, and vague responses, and gave structured feedback to aid in model refinement. On computational and STEM-related prompts, known correctness by Python-based reasoning and numerically-validated when possible. Adhered to the stringent quality control measures, such as consistency checks, compliance to the guidelines, and feedback loops of reviewers.

Worked on Generative AI data labeling project oriented towards assessing, grading, and refining AI text responses. These tasks included checking the outputs of the models in terms of facts, logical connections, depth of reasoning and compliance with the detailed instructions. Conducted RLHF-style assessments, where several model responses are compared to the same prompt and ranked in terms of quality, accuracy and clarity. Determined hallucinations, poor reasoning chains, and vague responses, and gave structured feedback to aid in model refinement. On computational and STEM-related prompts, known correctness by Python-based reasoning and numerically-validated when possible. Adhered to the stringent quality control measures, such as consistency checks, compliance to the guidelines, and feedback loops of reviewers.

2024 - 2024

Education

Work History