At AIPractix, we believe that building real-world AI products goes hand in hand with nurturing the next generation of AI engineers. We are proud to share a recent milestone: AIPractix successfully mentored student Ho Cam Truc, whose graduation thesis achieved excellent results with a strong focus on applied deep learning for visual search.
Graduation Thesis Overview
Title:
DESIGN AND IMPLEMENTATION OF A DEEP LEARNING–BASED IMAGE FEATURE REPRESENTATION SYSTEM FOR FASHION PRODUCT SEARCH FROM IMAGE QUERIES.
This thesis addresses a highly practical and increasingly important problem in modern e-commerce: how users can search for fashion products using images instead of keywords.
Rather than relying on traditional metadata-based search, the system enables users to upload an image and retrieve visually similar fashion items from a catalog, powered by deep learning–based feature representation.
Core Technical Contributions
Under the technical guidance of AIPractix, the thesis focused on building a production-oriented image feature representation and retrieval pipeline, including:
- Deep learning–based visual feature extraction for fashion items
- Embedding representation optimized for similarity search
- Image-to-image retrieval using vector similarity
- Scalable indexing strategy suitable for real-world catalogs
- Clear separation between model inference, feature indexing, and search serving
The system was designed not as a research prototype, but as a deployable foundation for real-world AI search engines.
Application to FeatSearch
The thesis outcomes were directly aligned with FeatSearch, AIPractix’s AI-powered search engine for structured catalogs.
Specifically, the work contributes to FeatSearch in the following ways:
- Image-based fashion search: users can find products using photos instead of text
- High-quality visual embeddings tailored for apparel and accessories
- Foundation for multimodal search, combining image queries with semantic text search
- Production-ready architecture, suitable for API-based and embedded deployments
This work strengthens FeatSearch’s vision of becoming a multimodal AI search engine for real commerce, where images, text, and structured attributes work seamlessly together.
Mentorship Philosophy at AIPractix
At AIPractix, mentorship goes beyond helping students “finish a thesis.”
Our approach emphasizes:
- Real-world problem selection, not purely academic topics
- System-level thinking, from model design to deployment
- Production constraints, including scalability and maintainability
- Bridging research and products, ensuring academic work creates practical value
Through this process, Ho Cam Truc gained hands-on experience with modern deep learning workflows that mirror real industry systems.
Looking Forward
We congratulate Ho Cam Truc on successfully completing this thesis with strong academic results and practical impact. This project demonstrates how applied AI education, when guided by real product experience, can produce solutions ready for industry adoption.
AIPractix remains committed to:
- Mentoring students in AI, Computer Vision, and Search Systems
- Collaborating with universities on applied AI research
- Translating academic work into real-world AI products like FeatSearch
If you are a student, researcher, or organization interested in applied AI mentorship or collaboration, feel free to connect with us at AIPractix.
AIPractix
Where Practical AI Powers Real Life
English 
Vietnamese