Academic • IR/Search • Visualization
About The Project
Designed an interactive tool to analyze approximate nearest-neighbor search behaviors by implementing HNSW and ACORN-1 indices, logging traversal steps, and projecting SBERT/GloVe embeddings into 3D via PCA for intuitive, semantic debugging. Exposed REST APIs for reproducible evaluation and demos.
Achievements
Built custom HNSW/ACORN-1 indices with traversal path logging
Unified 3D PCA space for cross-embedding comparison (SBERT/GloVe)
Real-time visualization for teaching, debugging, and interpretability
Empirical insight: ACORN-1 found more accurate neighbors at higher compute cost; HNSW took more steps but offered strong global jumps
More Projects
Links
sunny.sunho.park@gmail.com
