Dadang Irsyam

Key Skills

Software

Mindrift

Telus

Other

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Geospatial Tiled Imagery

Image

Text

Top Task Types

Fine Tuning

Object Detection

Prompt Response Writing SFT

Question Answering

Translation Localization

Freelancer Overview

I am a data and AI training specialist with over 2 years of experience working on diverse projects for global clients, including Outliers, Telus International, and other AI-focused platforms. My expertise spans data labeling, text annotation, and model training across multiple domains, with a strong focus on Natural Language Processing (NLP) and multimodal AI. Fluent in Indonesian, English, and Arabic, I have contributed to large-scale projects in machine translation, question answering, and conversational AI, ensuring high-quality datasets for finetuning and evaluation of Large Language Models (LLMs). Beyond text, I bring hands-on experience in image and video content review, as well as data quality assessment for AI-driven analytics. My background in linguistics and social science, combined with technical proficiency in AI workflows, enables me to bridge the gap between language, culture, and technology. I take pride in delivering precise, contextually relevant annotations and insights that empower models to perform in real-world scenarios with accuracy and fairness.

Labeling Experience

Paint-to-add-object Task

TelusImageObject Detection

This project focused on training generative AI models to realistically insert objects into visual scenes. The core task involved manually drawing precise masks (black overlays) on images to indicate where an imaginary object should be added. Annotators were required to fully cover the object and all related visual elements, including shadows, reflections, occlusions, and indirect lighting, ensuring there were no gaps, cracks, or missing areas in the mask. The annotation process used tools such as Brush (150px default), Lasso, and Pen, allowing for detailed and flexible masking. Each annotation was followed by a descriptive explanation of the object being "added"—including its appearance, size, placement, and interaction with the scene (e.g., lighting, perspective, surface contact). The project scaled across thousands of images featuring diverse environments (indoor, outdoor, day/night scenes). High-quality standards were maintained through reviewer feedback loops, cross-validation ch

This project focused on training generative AI models to realistically insert objects into visual scenes. The core task involved manually drawing precise masks (black overlays) on images to indicate where an imaginary object should be added. Annotators were required to fully cover the object and all related visual elements, including shadows, reflections, occlusions, and indirect lighting, ensuring there were no gaps, cracks, or missing areas in the mask. The annotation process used tools such as Brush (150px default), Lasso, and Pen, allowing for detailed and flexible masking. Each annotation was followed by a descriptive explanation of the object being "added"—including its appearance, size, placement, and interaction with the scene (e.g., lighting, perspective, surface contact). The project scaled across thousands of images featuring diverse environments (indoor, outdoor, day/night scenes). High-quality standards were maintained through reviewer feedback loops, cross-validation ch

2024

Make Up & Remove Object Task

TelusImageSegmentationClassification

This project involved synthetic image annotation for training AI models in object detection, removal, and inpainting. The primary task was to imaginatively place fictional objects within empty image spaces, apply accurate masking using annotation tools (e.g., brush, lasso), and provide detailed textual descriptions of the imaginary object along with precise removal and replacement instructions. The dataset comprised over 5,000 high-resolution images across various contexts (indoor, outdoor, urban, natural). High accuracy was maintained through iterative quality control checks, adherence to strict masking guidelines, and descriptive consistency. Reviewer feedback loops and cross-validation ensured that the annotations met both visual and contextual quality standards.

This project involved synthetic image annotation for training AI models in object detection, removal, and inpainting. The primary task was to imaginatively place fictional objects within empty image spaces, apply accurate masking using annotation tools (e.g., brush, lasso), and provide detailed textual descriptions of the imaginary object along with precise removal and replacement instructions. The dataset comprised over 5,000 high-resolution images across various contexts (indoor, outdoor, urban, natural). High accuracy was maintained through iterative quality control checks, adherence to strict masking guidelines, and descriptive consistency. Reviewer feedback loops and cross-validation ensured that the annotations met both visual and contextual quality standards.

2024

Data Collection

OtherImageData Collection

Conducted large-scale data collection and annotation of tourism-related documents (tickets, hotel bookings, itineraries) and finance billing records (invoices, receipts, transaction slips). Responsibilities included image review, classification, and metadata tagging to ensure accurate labeling for AI model training. Validated data quality by checking legibility, consistency, and categorization according to predefined taxonomies. Supported the development of OCR-based and image-recognition systems by delivering clean, well-structured datasets that improved model performance in document understanding, financial automation, and travel-related AI applications.

Conducted large-scale data collection and annotation of tourism-related documents (tickets, hotel bookings, itineraries) and finance billing records (invoices, receipts, transaction slips). Responsibilities included image review, classification, and metadata tagging to ensure accurate labeling for AI model training. Validated data quality by checking legibility, consistency, and categorization according to predefined taxonomies. Supported the development of OCR-based and image-recognition systems by delivering clean, well-structured datasets that improved model performance in document understanding, financial automation, and travel-related AI applications.

2024 - 2025

Text Fine Tuning for Indonesian and English

Scale AITextQuestion AnsweringText Generation

Contributed to the fine-tuning of a Large Language Model (LLM) for the Indonesian market, focusing on high-quality data annotation and linguistic accuracy. Tasks included creating and refining datasets for Question Answering, Text Generation, Text Summarization, Emotion Recognition, and Translation/Localization. Ensured cultural and contextual relevance of training data to improve model performance in natural and conversational Indonesian. Collaborated with global AI teams to review outputs, assess model accuracy, and provide feedback for continuous optimization. Delivered precise annotations that enhanced the model’s ability to handle nuanced queries, generate coherent responses, and maintain consistency across multilingual datasets.

Contributed to the fine-tuning of a Large Language Model (LLM) for the Indonesian market, focusing on high-quality data annotation and linguistic accuracy. Tasks included creating and refining datasets for Question Answering, Text Generation, Text Summarization, Emotion Recognition, and Translation/Localization. Ensured cultural and contextual relevance of training data to improve model performance in natural and conversational Indonesian. Collaborated with global AI teams to review outputs, assess model accuracy, and provide feedback for continuous optimization. Delivered precise annotations that enhanced the model’s ability to handle nuanced queries, generate coherent responses, and maintain consistency across multilingual datasets.

2024 - 2025

Education

Work History