Lucian-Alin Balint

Key Skills

Software

Micro1

Appen

Scale AI

Remotasks

Mindrift

Toloka

Labelbox

Mercor

Surge AI

Telus

Other

Top Subject Matter

AI Ethics

LLM Safety Evaluation

Fashion Image Generation & Annotation

Top Data Types

Text

Image

Document

Audio

Top Task Types

Object Detection

RLHF

Bounding Box

Transcription

Red Teaming

Evaluation/Rating

Data Collection

Question Answering

Text Generation

Text Summarization

Classification

Freelancer Overview

Interactive Prompt Specialist (Red Teaming). Core strengths include Mindrift and Gemini. AI-training focus includes data types such as Text and Image and labeling workflows including Red Teaming and Classification.

Labeling Experience

AI Safety & Red Teaming - LLM Evaluation

MindriftTextRed Teaming

Conducted red teaming and adversarial prompt testing on large language models to identify safety weaknesses and failure cases. Designed challenging prompts to evaluate instruction following, reasoning reliability, and policy compliance. Tested model responses for hallucinations, unsafe outputs, logical inconsistencies, and prompt injection vulnerabilities. Documented model weaknesses and contributed feedback to improve model robustness and alignment. Designed adversarial prompts targeting reasoning, safety, and instruction-following behavior Evaluated model outputs using structured rubric-based criteria Performed iterative prompt refinement to surface edge-case failures

Conducted red teaming and adversarial prompt testing on large language models to identify safety weaknesses and failure cases. Designed challenging prompts to evaluate instruction following, reasoning reliability, and policy compliance. Tested model responses for hallucinations, unsafe outputs, logical inconsistencies, and prompt injection vulnerabilities. Documented model weaknesses and contributed feedback to improve model robustness and alignment. Designed adversarial prompts targeting reasoning, safety, and instruction-following behavior Evaluated model outputs using structured rubric-based criteria Performed iterative prompt refinement to surface edge-case failures

2025 - Present

AI Model Evaluation Specialist – Outlier AI

OtherAudioTranscription

Evaluated AI-generated audio transcriptions for accuracy and alignment with spoken recordings. Reviewed model outputs to identify transcription errors, missing words, and contextual inconsistencies. Performed quality validation by comparing AI-generated transcripts against original audio inputs and documenting discrepancies. Applied structured evaluation guidelines to ensure consistency, accuracy, and reliable labeling across transcription validation tasks.

Evaluated AI-generated audio transcriptions for accuracy and alignment with spoken recordings. Reviewed model outputs to identify transcription errors, missing words, and contextual inconsistencies. Performed quality validation by comparing AI-generated transcripts against original audio inputs and documenting discrepancies. Applied structured evaluation guidelines to ensure consistency, accuracy, and reliable labeling across transcription validation tasks.

2025 - Present

AI Model Evaluation Specialist – Outlier AI

OtherImageBounding Box

Visual verification and bounding box annotation were performed on image datasets to assess AI object detection accuracy against ground-truth references. Tasks involved identifying objects and people, reviewing model-generated detections, and validating whether outputs matched the visual content correctly. Applied structured quality standards to ensure consistent object identification, accurate annotation, and reliable verification results across computer vision workflows. Maintained careful attention to detail while performing image review and quality control tasks.

Visual verification and bounding box annotation were performed on image datasets to assess AI object detection accuracy against ground-truth references. Tasks involved identifying objects and people, reviewing model-generated detections, and validating whether outputs matched the visual content correctly. Applied structured quality standards to ensure consistent object identification, accurate annotation, and reliable verification results across computer vision workflows. Maintained careful attention to detail while performing image review and quality control tasks.

2025 - Present

AI Product Prototype – Generative AI Image Fashion Imaging

OtherImageEvaluation RatingClassification

Developed generative AI pipelines for automated fashion image generation using advanced prompt engineering. Designed negative-constraint prompts to minimize visual artifacts and improve dataset quality. Tested scalable workflow integrations for AI-based image processing at production scale. • Created and annotated fashion datasets with high realism standards. • Eliminated anatomical anomalies through iterative prompt optimization. • Automated bounding box generation for object detection enhancement. • Leveraged workflow automation tools for dataset throughput scaling.

Developed generative AI pipelines for automated fashion image generation using advanced prompt engineering. Designed negative-constraint prompts to minimize visual artifacts and improve dataset quality. Tested scalable workflow integrations for AI-based image processing at production scale. • Created and annotated fashion datasets with high realism standards. • Eliminated anatomical anomalies through iterative prompt optimization. • Automated bounding box generation for object detection enhancement. • Leveraged workflow automation tools for dataset throughput scaling.

2025 - Present

AI Model Evaluation Specialist – Outlier AI

Micro1TextRLHF

Structured RLHF evaluation was performed on AI model outputs across text, audio, image, and reasoning tasks. Responses were compared and ranked using rubric-based criteria within human-in-the-loop evaluation workflows. Evaluated outputs for instruction following, factual accuracy, reasoning quality, safety compliance, and tone consistency. Executed structured reviews on math and logical reasoning chains. Conducted red teaming and adversarial prompt generation to surface model weaknesses and safety vulnerabilities.

Structured RLHF evaluation was performed on AI model outputs across text, audio, image, and reasoning tasks. Responses were compared and ranked using rubric-based criteria within human-in-the-loop evaluation workflows. Evaluated outputs for instruction following, factual accuracy, reasoning quality, safety compliance, and tone consistency. Executed structured reviews on math and logical reasoning chains. Conducted red teaming and adversarial prompt generation to surface model weaknesses and safety vulnerabilities.

2025 - Present

Education

Work History