Vincent Van Berchum

Key Skills

Software

Appen

Lionbridge

Scale AI

Internal/Proprietary Tooling

Top Subject Matter

No subject matter listed

Top Data Types

Text

Geospatial Tiled Imagery

Audio

Top Label Types

Action Recognition

Classification

Diagnosis

Evaluation Rating

Fine Tuning

Function Calling

Mapping

Question Answering

RLHF

Routing

Segmentation

Tracking

Transcription

Translation Localization

Freelancer Overview

I am an experienced AI trainer and data annotation specialist with a strong background in training, evaluating, and improving AI models for leading companies such as Outlier AI and Lionbridge. My work has spanned audio transcription, AI model testing, search engine evaluation, and a variety of data labeling tasks, particularly in natural language processing and multilingual content. With advanced computer skills, a keen eye for detail, and a perfectionist mindset, I consistently deliver high-quality training data for AI and machine learning projects. My expertise also includes translation, localization, and editing across multiple platforms, which has given me a deep understanding of linguistic nuances and data quality standards essential for effective AI development.

Labeling Experience

Outlier AI project

Internal Proprietary ToolingTextClassificationRLHF

A single prompt and a set of two responses to that same prompt are presented to me. My job is to classify the responses, rate and evaluate them according to specific dimensions (mostly linguistic in nature) and provide feedback. Finally, I rate which of the responses satisfies the user intent in the best way, according to the linguistic dimensions, and provide feedback on why that is the case. This is all done according to extensive manuals that detail the way that ratings and evaluations should be done, based on specific conditions laid out in the manuals.

A single prompt and a set of two responses to that same prompt are presented to me. My job is to classify the responses, rate and evaluate them according to specific dimensions (mostly linguistic in nature) and provide feedback. Finally, I rate which of the responses satisfies the user intent in the best way, according to the linguistic dimensions, and provide feedback on why that is the case. This is all done according to extensive manuals that detail the way that ratings and evaluations should be done, based on specific conditions laid out in the manuals.

2025

Lionbridge AI project

LionbridgeTextClassificationQuestion Answering

During this project I would rate and classify both prompts and responses from an AI model across various dimensions, including dimensions that focus on linguistics, harmful content and jailbreak attempts. The ratings and classifications were based on conditions laid out in extensive manuals. In addition to that, I would take responsibility for the quality assurance (QA) and feedback on the work of other AI trainers for the same project.

During this project I would rate and classify both prompts and responses from an AI model across various dimensions, including dimensions that focus on linguistics, harmful content and jailbreak attempts. The ratings and classifications were based on conditions laid out in extensive manuals. In addition to that, I would take responsibility for the quality assurance (QA) and feedback on the work of other AI trainers for the same project.

2024 - 2025

Lionbridge / Google project

LionbridgeAudioClassificationQuestion Answering

Project for Google Assistant voice recognition and transcription on mobile phones through Lionbridge. I would listen to audio from users interacting with Google Assistant, transcribe it, classify whether the user intended for Google Assistant to activate and whether their questions were answered adequately by the resulting Google search. The transcribing and classifications were done according to Google's extensive manuals for this specific project.

Project for Google Assistant voice recognition and transcription on mobile phones through Lionbridge. I would listen to audio from users interacting with Google Assistant, transcribe it, classify whether the user intended for Google Assistant to activate and whether their questions were answered adequately by the resulting Google search. The transcribing and classifications were done according to Google's extensive manuals for this specific project.

2017 - 2018

Appen / Microsoft Bing project

AppenGeospatial Tiled ImageryAction RecognitionTracking

Data labeling project for the Microsoft Bing search engine (and their Maps function) through Appen. My work consisted of labeling prompts which the search engine would recognize as user requests for a map of a location that the user is searching for and evaluating whether the Bing Maps result of the prompt complies adequately with the user request. Routes and accuracy of maps were rated while also noting occurrences where Bing would not recognize a map request while it should (and vice versa). I carried out this work with the help of extensive manuals that included conditions for rating and evaluating prompts and search results that include maps.

Data labeling project for the Microsoft Bing search engine (and their Maps function) through Appen. My work consisted of labeling prompts which the search engine would recognize as user requests for a map of a location that the user is searching for and evaluating whether the Bing Maps result of the prompt complies adequately with the user request. Routes and accuracy of maps were rated while also noting occurrences where Bing would not recognize a map request while it should (and vice versa). I carried out this work with the help of extensive manuals that included conditions for rating and evaluating prompts and search results that include maps.

2013 - 2013

Education

Work History