About

I'm a machine learning engineer working on computer vision, multimodal models and large language models. I build systems that understand images and video, and language models that answer from evidence rather than from memory.

I care most about reliability: whether a model still works outside its benchmark, and whether a person can check what it says. That shapes how I build models and how I evaluate them.

I'm open to research collaborations in these areas. If you're working on something related, I'd be glad to hear about it.

Research interests

  • Computer Vision
  • Large Language Models
  • Multimodal & Vision-Language Models
  • Medical & Biomedical Imaging
  • Trustworthy AI

Across all of these, I'm drawn to one question: do models still behave reliably once they leave the benchmark?

News

  • 2026 Paper

    Submitted Joint Prediction of UHPC Properties Using a Feature-Tokenizer Transformer to Applied AI Letters. It's under review.

  • Jan 2025 Career

    Joined Accelx Inc. as a Machine Learning Engineer.

  • Jan 2021 Thesis

    Graduated from AUST with a thesis on arrhythmia classification using 2-D CNNs.

Selected projects

All projects →

  1. 1. Overcooked AAR — evaluating video-language models

    GitHub repository ↗

    Can a Video LLM review a human team's play as well from video as from the game's event log? I had Qwen2.5-VL and Qwen3-VL write after-action reviews of real two-player Overcooked games both ways, then checked every timestamped claim against the log. On 15 episodes, claims from the log were 67–93% precise against 22–43% from video, and the video timestamps were no better than random.

  2. 2. DocuVision — key–value extraction from scanned documents

    GitHub repository ↗

    DocuVision turns scanned forms, PDFs and phone photos into structured key–value pairs. I fine-tuned LayoutLMv3-large and built a learned question→answer linker. On FUNSD, pair F1 rose from 0.23 to 0.74 with ground-truth words, and from about 0.15–0.17 to 0.43–0.46 end to end on the raw scans with OCR. I released the model on Hugging Face.

  3. 3. Sentinel — tracking-based video anomaly detection

    GitHub repository ↗

    Sentinel watches a video stream and raises alerts for falls, loitering, abandoned objects, crowding and wrong-way motion. I combined YOLOv8 detection and BoT-SORT tracking with a rule engine and deduplicated alerts, and the whole pipeline runs at 19.7 FPS on an RTX A5000.

  4. 4. DocuRoute — routing between retrieval and text-to-SQL

    GitHub repository ↗

    DocuRoute answers questions about SEC 10-K filings with citations. An LLM router I built sends each question to hybrid retrieval or to sandboxed text-to-SQL. In my evaluation harness, a general-domain reranker lowered faithfulness from 0.93 to 0.77, so I kept plain hybrid retrieval as the default.

Let's collaborate

I'm open to research collaborations, especially on vision, video, or language models. A short email about what you're working on is the best way to start.

Get in touch