Artificial Intelligence & Deep Learning

Investigating core representations, attention rewiring, and scalable ML inference systems.

shoppin' Fashion Search & Model Training

Building a fashion discovery search lens that maps snap-to-search items compared across 10 websites. Designed and benchmarked 10-12 image search methods with full performance metrics.

• Model Training: Ran model training on extensive image sets to construct search matching.
• Cosplay Search Node: Designed a cosplay search extension feeding characters to extract and match coordinates for clothes across the web.
• Multimodal Vector Ingestion: Managed large-scale product catalogs matching 3-4 crore catalog items using CLIP embedding models, deployed on AWS SageMaker pipelines.

Transformer Optimizations

Adapted attention mechanics inside transformer-based architectures (DistilBERT, GPT-2) to prioritize technical semantics and vulnerability text reports, improving cyber-vulnerability exploit prediction accuracy to 85.6%.

Relevant Resources

Read about our published research at the Publications Page or examine my open-source packages at the Open Source Page.