RAG pipeline that vectorizes PDFs and performs semantic search before generating grounded LLM responses.
Multimodal ML model estimating eVTOL pilot stress levels from 7 biometric modalities. Ground truth via NASA TLX.
Compressed a 72M-parameter garment try-on model onto Jetson Nano 4GB via quantization, pruning, and distillation.
Transformer-based joint prediction model trained on 200 days of FAA SWIM trajectory data from KSEA and KEWR.
CNN classifier on AffectNet (291K images, 8 labels) achieving ~70% validation accuracy with full error analysis.