Mohita Chowdhury

I am currently working as the Lead AI Research Engineer at Ufonia where I drive the work to improve the performance and intelligence of our autonomous telemedicine clinical assistant. I am currently leading strategy and R&D for the integration of large language models into our regulated clinical product.

I graduated with a Master's degree in Computer Science from the University of Oxford in October 2020. For my Master's thesis, I worked with the Visual Geometry Group (VGG) on developing a zero-shot human-object interaction framework using visual, semantic and spatial modalities. This work was supervised by Prof Phil Blunsom and Dr Vicky Kalogeiton.

Previously, I've worked as a Research Assistant at the Hardware and Embedded Systems Lab at Nanyang Technological University, Singapore under Prof Thambipillai Srikanthan and Dr Nirmala Ramakrishnan. I developed a low-cost algorithm to prune headlight reflections from foreground detectors for robust nighttime traffic surveillance.

I've also interned at the Samsung Innovation Lab at Indian Institute of Technology Delhi (IIT Delhi) under Prof Brejesh Lall. During this internship, I developed a framework for robust object segmentation using stereo images.

I have a Bachelor's degree in Electronics and Communication Engineering from National Institute of Technology Karnataka, Surathkal, India. For my BTech thesis, I worked under the guidance of Prof Sumam David on Copy-Move Forgery Detection in images.

My interests are broadly in machine learning, natural language processing and computer vision. I have worked on problems in the domains of object segmentation, copy-move forgery detection, foreground detection, multimodal interaction detection, and conversational AI.

Email  /  Resume  /  Google Scholar  /  Github  /  LinkedIn

This study uses data collected from 120 patients using an autonomous telemedicine clinical assistant that asks symptom-based questions to elicit patient concerns and allows patients to ask questions about their post-operative recovery. With the help of ophthalmologists, we examine the safety and appropriateness of the responses generated to patient questions by the popular LLM by OpenAI, ChatGPT.

Developed a self-supervised learning-based error-simulator to generate synthetic but realistic ASR-like errors to train ASR error correction models.

The study uses data collected from 191 patients using an automated telephone platform that asks symptom-based questions to elicit postoperative concerns. We compare the ability of different machine learning techniques to understand patients' descriptions of their symptoms during a cataract surgery follow-up with Dora.

Developed a SIFT and GLCM-based approach to leverage textural similarity as a means to detect copy-move forgeries in images, even under the influence of multiple geometrical transformations like scaling and rotation.

Website Template