For employers

Hire this AI Trainer

Sign in or create an account to invite AI Trainers to your job.

LoginCreate account
Invite to Job
Muhammad Ihsan Hutri Roswandi

Muhammad Ihsan Hutri Roswandi

AI Trainer

Indonesia flagKota Bandung, Indonesia
$8.00/hrIntermediateClickworkerLabelboxMindrift

Key Skills

Software

ClickworkerClickworker
LabelboxLabelbox
MindriftMindrift
OneFormaOneForma
Scale AIScale AI
TelusTelus
Internal/Proprietary Tooling
Other

Top Subject Matter

Clothing imagery
Indonesian Prompt Generator
Response Writing

Top Data Types

AudioAudio
ImageImage
TextText

Top Task Types

Audio Recording
Bounding Box
Data Collection
Prompt Response Writing SFT
Text Generation

Freelancer Overview

With extensive experience in data labeling and AI training data, I have contributed to impactful projects at Pareto AI, Outlier AI, and TELUS Digital. At Pareto AI, I enhanced online product presentations by reviewing and categorizing product and lifestyle images, crafting detailed captions, and applying bounding boxes for precise labeling. At Outlier AI, I improved AI performance by crafting tailored prompts, evaluating AI responses using the Likert scale, and rewriting outputs for clarity and relevance. These roles honed my skills in meticulous analysis, data structuring, and effective communication. At TELUS Digital, I assessed personalized ads for relevance and quality, ensuring they aligned with user preferences. My ability to handle sensitive data with strict confidentiality, coupled with constructive feedback to marketing teams, demonstrated my commitment to delivering high standards. These experiences underscore my expertise in data annotation, quality evaluation, and AI optimization, making me adept at driving improvements in AI-driven projects.

IntermediateIndonesianEnglishSundanese

Labeling Experience

AI Content Post Editor on CGP Project

OtherTextTranslation Localization
In the CGP Project at LILT, I refined AI-generated content across formats such as SMS, email, and social media posts. My task was to ensure that each piece adhered to natural language usage, appropriate tone, and grammatical accuracy, making it suitable for real-world communication. The content was localized to reflect Indonesian language norms and cultural context. I managed and post-edited around 300 texts per week, adjusting stylistic tone (formal/informal) based on content type. This work helped improve the AI’s ability to generate locally relevant, authentic, and user-friendly output, contributing to better customer experience in multilingual communication tools.

In the CGP Project at LILT, I refined AI-generated content across formats such as SMS, email, and social media posts. My task was to ensure that each piece adhered to natural language usage, appropriate tone, and grammatical accuracy, making it suitable for real-world communication. The content was localized to reflect Indonesian language norms and cultural context. I managed and post-edited around 300 texts per week, adjusting stylistic tone (formal/informal) based on content type. This work helped improve the AI’s ability to generate locally relevant, authentic, and user-friendly output, contributing to better customer experience in multilingual communication tools.

2025

AI Trainer on Multispeaker Convo Audio Project

OtherAudioTranslation LocalizationAudio Recording
This project involved recording and annotating natural-sounding conversations between 2 to 5 speakers, simulating real-life interaction dynamics. The conversations covered a range of topics and tones, allowing the AI to learn from diverse dialogue structures, interruptions, informal expressions, and speaker overlaps. I ensured high audio quality using professional recording equipment and adhered to strict technical standards. Post-recording, I segmented speaker turns and labeled conversational scenarios to reflect the emotional tone, setting, and intent behind each interaction. This detailed labeling enabled the AI model to better detect speaker boundaries, interpret human intent, and respond in a more contextually aware and adaptive manner during conversations.

This project involved recording and annotating natural-sounding conversations between 2 to 5 speakers, simulating real-life interaction dynamics. The conversations covered a range of topics and tones, allowing the AI to learn from diverse dialogue structures, interruptions, informal expressions, and speaker overlaps. I ensured high audio quality using professional recording equipment and adhered to strict technical standards. Post-recording, I segmented speaker turns and labeled conversational scenarios to reflect the emotional tone, setting, and intent behind each interaction. This detailed labeling enabled the AI model to better detect speaker boundaries, interpret human intent, and respond in a more contextually aware and adaptive manner during conversations.

2025 - 2025

AI Trainer on Xylophone Grassland Project

Internal Proprietary ToolingAudioClassificationTranslation Localization
In the Xylophone Grassland Project, I created and recorded prompts across a broad range of categories, each designed to simulate different real-world conversational scenarios. These prompts varied in tone, style, and complexity, and were used to help train AI models to understand diverse human communication patterns. I also added background sounds like ambient noise from public transportation or cafes to enhance the realism of the audio. Each recording was carefully labeled with contextual metadata, including environmental conditions, speaker tone, and category type. These annotations helped the model learn to distinguish and respond appropriately in different acoustic and social environments. The work was instrumental in improving the robustness of conversational AI across varied use cases.

In the Xylophone Grassland Project, I created and recorded prompts across a broad range of categories, each designed to simulate different real-world conversational scenarios. These prompts varied in tone, style, and complexity, and were used to help train AI models to understand diverse human communication patterns. I also added background sounds like ambient noise from public transportation or cafes to enhance the realism of the audio. Each recording was carefully labeled with contextual metadata, including environmental conditions, speaker tone, and category type. These annotations helped the model learn to distinguish and respond appropriately in different acoustic and social environments. The work was instrumental in improving the robustness of conversational AI across varied use cases.

2025 - 2025

AI Trainer on Beryl Project

Internal Proprietary ToolingImageBounding BoxQuestion Answering
The Beryl Project aimed to train an AI agent to better understand and interact with user interface elements by simulating real-world screen-based tasks. My main responsibilities included creating Locate queries by drawing bounding boxes around key UI components, allowing the agent to visually recognize and respond to interactable elements. Additionally, I developed VQA (Visual Question Answering) queries based on screen content, providing clear text answers that required contextual comprehension and accurate visual reasoning. This project focused on refining the agent’s reasoning skills, particularly for non-obvious, multi-step queries. The quality of work was assessed based on the clarity, accuracy, and consistency of both bounding box placements and answer relevance. High standards were maintained throughout the project, contributing to improved AI performance in understanding complex UI structures and executing precise user-based commands.

The Beryl Project aimed to train an AI agent to better understand and interact with user interface elements by simulating real-world screen-based tasks. My main responsibilities included creating Locate queries by drawing bounding boxes around key UI components, allowing the agent to visually recognize and respond to interactable elements. Additionally, I developed VQA (Visual Question Answering) queries based on screen content, providing clear text answers that required contextual comprehension and accurate visual reasoning. This project focused on refining the agent’s reasoning skills, particularly for non-obvious, multi-step queries. The quality of work was assessed based on the clarity, accuracy, and consistency of both bounding box placements and answer relevance. High standards were maintained throughout the project, contributing to improved AI performance in understanding complex UI structures and executing precise user-based commands.

2025 - 2025

AI Trainer on Gold Project

Internal Proprietary ToolingVideoSegmentationText Generation
The Gold Project focused on enhancing AI's ability to generate and understand longer video content, specifically cartoon clips. My tasks involved creating detailed captions and scene descriptions for 3-second video segments, capturing key visual and narrative elements. I provided descriptions of characters, emphasizing their visual traits and actions within each scene to help train the AI model to generate coherent video content. These captions also included integrated scene content and character movement to ensure comprehensive understanding. The project required high accuracy in describing the scenes, with minimal revisions necessary. The quality was measured by how well the AI model could use the generated captions to improve its video content generation. The success rate was consistently high, with task performance meeting the project’s standards, ensuring clear, informative, and contextually relevant descriptions.

The Gold Project focused on enhancing AI's ability to generate and understand longer video content, specifically cartoon clips. My tasks involved creating detailed captions and scene descriptions for 3-second video segments, capturing key visual and narrative elements. I provided descriptions of characters, emphasizing their visual traits and actions within each scene to help train the AI model to generate coherent video content. These captions also included integrated scene content and character movement to ensure comprehensive understanding. The project required high accuracy in describing the scenes, with minimal revisions necessary. The quality was measured by how well the AI model could use the generated captions to improve its video content generation. The success rate was consistently high, with task performance meeting the project’s standards, ensuring clear, informative, and contextually relevant descriptions.

2025 - 2025

Education

P

Padjadjaran University

Bachelor of Applied Science in Digital Marketing, Digital Marketing

Bachelor of Applied Science in Digital Marketing
2021 - 2025

Work History

No Work History added yet

Muhammad Ihsan H. hasn’t added any Work History to their OpenTrain profile yet.