ShowHue is a computer vision startup that provide detection and generation model API for eCommerce and marketing platform, generating marketing-ready image without prompts input and photo edits.
As a co-founder in ShowHue, I was in charge of product design and growth strategies. I managed 10+ person cross-functional team, including science, design and business, ensuring communicating with internal teams to deliver customers' expectations and needs.
Milestones
- Featured on Taiwan AI Ecosystem Map for 3 years
- Won the 2019 Asia Open Data Challenge Champion
- Forged partnerships with over 100 enterprises
- Achieved a monthly growth rate of 600% for AI image-generating SaaS service
- Secured over $1M in grants for AI startup from the Taiwan government
Impact & Contributions
Enhancing Image Generator API Performance
Analyzed data delivery metrics and conducted user research to identify areas for fine-tuning and improvement, resulting in enhanced performance and reliability of our image generation API
- Improved image quality from 26.98% to 62.75% by integrating LLMs (Large Language Models) and object detection models
- Achieved a 5X increase in API documentation views within 8 weeks
Human Eval System for Image Generation Model
Conduct thorough evaluations of how prompting techniques affect results align with project requirements
- Conducted beta testing on 11 model versions for a new model capability using a robust evaluation process
- Standardized visual inspection using the perceptual quality method and automatically generated 12,000+ testing images
- Led the creation of evaluation datasets for potential risks and harms of image generation models
RAG for Enhanced Product Imagery
Implemented a retrieval-augmented generation (RAG) approach with a curated image database to automatically generate visually appealing and contextually relevant product images for marketing and e-commerce applications.
- Curated a high-quality image database by filtering 1 million images down to 611k images
- Developed an automated image classification system to organize the 611k images into 5 distinct categories
Automated Image Data Curation and Collection
Implemented robust data strategies, including automated web scraping and AI filtering. Ensuring high-quality data delivery for e-commerce and marketing, and oversaw data operations to ensure policy compliance.
- Collected 1 million images from 6+ target platforms using web scraping, ensuring policy-compliant data collection
- Executed data cleaning and engineering through an automated filtering system that achieved an 81.56% success rate in preparing data for AI model training, using classification, object detection, and SVM algorithms
- Established data pipelines and data management systems to track data sources, monitor data standardization
Scalable Annotation Platform
Developed an annotation tool with labeling interfaces to asses labeled data for data annotation workflows, labeling processes and model training preparation.
- Successfully managed over 500 applicants for the annotation project
- Managed an engineering team of 20+ annotators, collecting feedback from AI trainers, achieving 100,000+ fashion-labeled images
- Defined comprehensive labeling guidelines, writing instructions and success criteria to ensure data quality and data consistency
Secure AI Integration API Development
Developed a secure and testable token-based API with clear documentation to enable B2B clients to seamlessly integrate AI image generation capabilities into their platforms.
- Deployed 5 image generation APIs
- Tested 17+ AI models to refine API output and content policies
- Developed 11 API endpoints aligned with the PRD
- Provided operational support and document maintenance through a ticket system