Welsh Audio Specialist - Freelance AI Trainer Project

Project Overview

We are sourcing independent Audio Evaluation Specialists for an AI benchmark evaluation project assessing advanced agentic audio models. As AI models increasingly handle complex workflows in this domain—specifically real-world customer support scenarios like flight bookings, financial services, and telecommunications—their accuracy relies entirely on robust, expert-crafted training data. The objective of this project is to autonomously produce high-quality evaluation tasks through simulated interactions, audit conversational AI outputs, and generate clean, reliable datasets to optimize model performance.

Project Deliverables & Scope

Operate autonomously to design complex evaluation frameworks and provide structured training data. Expected deliverables include:

Role-Play Scenario Execution: Creating and executing complex, role-play-based evaluation scenarios that simulate realistic customer service interactions across travel, finance, and technical support domains. Model Performance Auditing: Evaluating AI model performance across standardized qualitative and quantitative metrics, focusing strictly on task completion accuracy, conversational naturalness, and audio comprehension. Technical Metric Evaluation: Assessing the model's basic computer programming literacy, including its understanding of JSON structures, functions, methods, and ability to reason about structured data within a support context. Representative Dataset Generation: Contributing to the development of diverse, high-quality audio datasets that accurately reflect real customer expectations for clarity, efficiency, and natural conversational flow.