Mercor is an AI-powered, project-centric platform that connects elite, vetted, global talent with top-tier AI labs and enterprises for specialized, remote contract roles in software engineering, law, data, consulting and other highly-specialized domains.

My work and roles here vary from project-to-project on which I work when they become available. Lately, I have been working as an Evaluator, Annotator, Prompt Writer and Reviewer for supervised fine-tuning (SFT) datasets designed to train Large Language Models. (100% remote). Specifically, I -

• Source high-quality textual materials used as "true-source" documents for training the model in specific domains.
  
  
• Write and calibrate complex prompts and answer keys which are fed to the models during their training.
  
  
• Run diagnostics and augment generated Golden Responses to improve models' responses for formatting, factual accuracy, content groundedness, alignment with answer keys, and with the style guides they had been fed.

With a rapidly-gained valuation (currenly at $10 billion) and undisputed industry trust, currently Mercor works with over 50,000 higly-specialized professionals who contribute to training large language models by working in coding, data collection, annotation, labeling, validation, evaluation and other AI/ML-specific tasks. The platform is designed for skilled experts rather than beginners, targeting the top 10%-20% of talent to drive AI model improvement. (also paying significantly higher rates than competitors averaging a pay rate of $95 per hour).

TransPerfect/DataForce is a global data collection, labeling and AI data gathering platform which allows users to design and manage data structuring and categorization of audio, image, video annotation, and transcription for customer experiences.

During my work there, I managed four AI data collection and structuring projects where, in addition to acting as a moderator, I managed several on-site projects assigned to DataForce's New York City office.

Some projects' scopes (where I moderated the on-site data collection sessions) were to collect data through actionable intelligence studies which assessed human-machine interaction patterns, and to create metadata and structured data tables which were later used in the development of machine learning algorithms. In other projects the data collection focussed on recording and classifying (guided and free) conversations between participants who were speaking different dialects of the English language, and/or with a specific accent such as Spanish, Chinese or East Indian.

These projects, and their respective generated/collected data, were part of language analysis and interpretation (speech recognition), and were utilized to validate trained datasets used in the development of new AI technologies embedded in human intelligence applications, speech recognition, and machine learning.

At the end of each day I loaded data to dedicated project servers (S3/AWS) for editors and QA reviews, and created/submitted daily workflow reports. Some of the collected data were used in the development of user demographic profiles based on which I built targeted email lists. Between data collection sessions I used these lists to advertise, recruit and invite selected participants to take part in these projects.