Research index
Current investigations, collaborative systems work, and earlier studies. Each project page describes the question, contribution, evidence, and remaining limitations.
I study how candidate access, evidence selection, and reading affect multi-hop answers. Preliminary findings motivate further comparisons with stronger retrieval and scoring methods.
I investigate how speech-predictable facial motion relates to rendered articulation. Following a negative predictive pilot, RIFT-M focuses on measurement reliability.
I am building a verified self-training pilot to study model expansion. Reliable training targets and recovery checks come before testing the growth hypothesis.
Collaborative MILab and industry work on speech-to-avatar integration, distributed inference, and Jetson deployment. Niloy and I are exploring a manuscript direction from the deployed PersonaPlex–IMTalker pipeline.
A preliminary browser workbench developed with aerospace practitioner input: STL inspection, custom VLM analysis, and OpenFOAM case preparation. CFD simulations remain future work.
Our completed four-person undergraduate thesis combines pretrained perception with SQL, multi-aspect vector retrieval, and temporal memory. I am considering architectural renovation and renewed evaluation.
I tested affine covariance transport in Gaussian avatars. Geometric differences did not support the required association with rendering residuals across three identities, so I set the hypothesis aside.
A C++ prototype for dependency and cycle analysis, pragma injection, and OpenMP-based CPU and target execution.
A Peekaboo/DINO teacher and UKAN student with mask-supervised and teacher–student objectives.