Research Assistant
с 05.2025 - По настоящий момент |Nepal Applied Mathematics and Informatics Institute for research
Pytorch, OpenCV, MMCV
Research and apply State of the Art AI techniques on medical imaging, specifically on Medical Image Segmentation, Out of Distribution Detection and Federated Learning
Computer Vision Research Engineer
с 06.2025 - По настоящий момент |Redev AI
Pytorch, OpenCV, Hugging Face
Experiment with State of the Art VLMs for Automatic Annotation, Training Detection and Classification Models
Software Coordinator
с 06.2024 - По настоящий момент |LOCUS
CRM
● Managed software-related events for LOCUS 2025, including the Software Fellowship, Hackathons, and AI Competitions.
● Oversaw and led various software projects under LOCUS and the LOCUS Open Source Team (LOST).
● Worked on screening and guiding software projects for the exhibition at LOCUS, Nepal's largest technological festival
Vice President
08.2019 - 02.2020 |St. Xavier's College Computer Club
CRM
● Organized club events and competitions, including the CSP Olympiad, Web Development Competition, and the SXC Computer Festival.
Computer Vision
Projects: Reconstruction of Heritage Structures
Python, PyTorch, 3D Gaussian Splatting, gsplat, Computer Vision, 3D Reconstruction
● Researched dataset capturing methodologies for the 3D reconstruction of large heritage structures.
● Utilized state-of-the-art Gaussian Splatting techniques (Hierarchical Gaussian Splatting, CityGaussian, 3DGS, gsplat) to reconstruct and compare 3D models for their effectiveness in this task.
● Investigated the impact of using masks and bilateral grids on the reconstruction quality of 3D models.
● Researched and worked on methods to build virtual walkthroughs of heritage structures.
Computer Vision
Projects: Virtual Try-On of Shoes
Python, PyTorch, SAM (Segment Anything Model), Gaussian Splatting, Snapchat LensStudio, AR, Image Segmentation, Dataset Curation
● Collected over 100 videos of shoes and sampled 30 images per video to create a dataset of 3000 shoe images from various orientations, resolutions, lighting conditions, and backgrounds.
● Created an automatic annotation pipeline with SAM to produce a large-scale shoe segmentation dataset.
● Trained different segmentation models on the created dataset.
● Utilized Gaussian Splatting to generate 3D models of the shoes from segmented images.
● Built an AR application using Snapchat's LensStudio to overlay the generated 3D shoe models onto a user's foot.
Computer Vision
Projects: Cattle Muzzle Pattern Matching
Python, PyTorch, Siamese Networks, ViT, ResNet, EfficientNet, Grad-CAM
● Researched existing methodologies, pre-processing techniques, models, and metrics for pattern matching.
● Trained a Siamese Model with various backbones (ViT, ResNet, EfficientNet) and compared their performance.
● Utilized Grad-CAM visualizations to interpret model results.
● Experimented with different datasets, models, and pre-processing techniques, performing ablation studies.
Computer Vision
Projects: SCOPE - Semantic Captioning for Optimized Photo Exploration
Python, PyTorch, Image Captioning, ViT, DeiT, GPT-2, BERT, Model Quantization, Knowledge Distillation, Embeddings
● Researched deep learning methods for Image Captioning.
● Experimented with combinations of different Vision Encoders (ViT, DeiT) and Text Decoders (GPT-2, distilGPT).
● Evaluated the performance of large and small models under time-constrained training scenarios.
● Researched downsizing methods for large models (Knowledge Distillation, Quantization) to enable edge-device deployment.
● Quantized GPT-2, BERT, and ViT models to reduce their space and compute requirements.
● Developed a system that generates image captions, processes them through BERT to create embeddings, and saves them as metadata to enable text-based image searches.
Computer Vision
Project
Python, PyTorch, CNNs, Medical Image Analysis, Image Classification, Dataset Curation, Transformers, Deep Learning Fundamentals
Personal Project:
● Trained classification models with different CNN backbones to classify fundus images into normal, three disease categories (Diabetes, Glaucoma, Cataract), or other.
● Analyzed model performance and dataset, proposing measures to increase accuracy.
● Searched for additional data and wrote transformation scripts to align the new dataset with the old one.
● Identified and corrected inconsistencies in the dataset labels by writing correction scripts.
Other Projects:
● Mahakabi: Explored generative models for creative text generation.
● Vision Transformer Implementation: Built a Vision Transformer model from scratch.
● Anode: A novel project concept and implementation.
Teaching Assistant
01.2025 - 01.2025 |NAAMII
CRM
● Volunteered as a Teaching Assistant, supporting the Foundational Models lab and guiding participants through a project on SimCLR.
Lead, Children in Technology / Lead, Software Fellowship
07.2024 - 11.2024 |LOCUS
CRM
● Led a team of six to educate high school students on internet safety and technology.
● Conducted sessions in eight schools across four districts of Nepal.
● Organized a 10-day workshop on software fundamentals for over 250 beginner IT students.
● Led a team of over 30 volunteers in designing the syllabus and preparing teaching materials covering Web Design, Frontend, Backend, Deployment, and Data Science.
● Oversaw sponsorships, scheduling, and logistics for the event.
Assistant, Seminar on Generative AI
06.2023 - 06.2023 |IT Club
CRM
● Assisted participants in understanding generative model concepts and guided them through coding and executing the examples.
