Multimodal AI · Video understanding · Inference systems
I build AI that can perceive and reason about the world more like we do.
My research spans multimodal perception, video understanding, reasoning, reinforcement-learning post-training, and efficient AI systems.
Research focus
About
I develop models that connect perception with higher-level understanding. My past work focused on video understanding and visual perception, and I'm now expanding into multimodal models that follow events through time, ground language in visual evidence, and learn coherent reasoning strategies.
I recently completed my Ph.D. in Computer Science at Dartmouth College. I work across video understanding, visual grounding, reinforcement-learning post-training, and transparent reasoning evaluation.
I'm also interested in high-performance AI infrastructure and efficient inference systems, including continuous batching, multimodal prefix caching, and native model serving on Apple Silicon that reaches hundreds of tokens per second.
My work has appeared at CVPR, ICCV, ECCV, ICML, WACV, and other leading venues, as well as in PLOS Digital Health, the Journal of Pathology Informatics, Frontiers in Medical Technology, and other journals.
I earned my Ph.D. at Dartmouth College, advised by SouYoung Jin, and founded Wiqonn, an AI lab growing ambitious AI work in Latin America.
Selected research
Latest · Multimodal reasoning
A 6,372-instance benchmark that checks every intermediate step. Across 20 models, none maintained more than 60% step accuracy in the correct order; Causal Process Reward improves step-level consistency by 32% without manual annotations.
Introduces native MLX inference with continuous batching and multimodal prefix caching, reaching up to 525 tokens per second and 28x faster repeated image queries on M4 Max.
Uses instruction-guided channel modulation to improve fine-grained visual grounding across three MLLM families and 12 benchmarks with less than 1% additional FLOPs.
Combines AI biomarkers from chest radiographs, clinical notes, and ECG signals to study pneumonia mortality using routinely collected hospital data.
Generates talking faces directly from first-person text in a one-stage pipeline, without requiring audio, landmarks, or pose during inference.
Introduces multi-layer learnable attention masks that prioritize informative tokens across transformer layers while reducing redundant computation.
Improves language grounding in long videos by identifying describable windows and pruning irrelevant segments before matching a query.
Introduces Bladder4Net to classify whole-slide histopathology images into low-risk and high-risk bladder cancer groups for prognosis support.
Builds a weakly supervised CT pipeline for COVID-19 classification, lesion segmentation, and quantification from scan-level labels.
Analyzes the training and testing gaps between proposal, classification, and localization modules in event detection pipelines, then introduces practical fixes.
Uses object and scene semantics to generate class-specific action proposals and prune unrelated activities in real time on ActivityNet.
Open source
My research and tools first, then contributions to open-source communities — video, geospatial, and LLM infrastructure.
Building Wiqonn and researching multimodal perception, video understanding, and efficient inference.
Get in touch