About
Software Engineer with a Passion for AI/ML Applications Development and its Deployment
Welcome to my digital domain, where Artificial Intelligence (AI) is not just a subject of fascination but a canvas for innovation. My journey in AI was kindled by Andrew NG's 2008 CS229 Machine Learning Stanford lecture. Since 2017, I've been deeply engrossed in exploring the enigmatic world of AI, particularly how machines perceive and influence our world.
At the heart of my current exploration is computer vision and machine learning, realms where I am passionately investigating how machines understand visual data. My research is geared towards developing architectures that learn efficiently from minimal data, employing advanced methodologies like few-shot learning. This pursuit aligns with my overarching goal of contributing to the sphere of artificial general intelligence.
A significant aspect of my expertise lies in deploying machine learning models in the cloud. I am fascinated by and skilled in creating scalable, serverless applications, drawing inspiration from industry pioneers like Meta and Google. Their proficiency in developing systems that seamlessly serve millions of users worldwide serves as a benchmark in my professional aspirations.
Technical Expertise and Tools:
- Core Areas: Deep Learning, Machine Learning, Computer Vision, Data Science, and ML Model Deployment in the Cloud.
- Programming Languages:Proficient in Python, Knowledgeable in Java, C, C++, HTML/CSS, and Bash.
- Tools & Technologies: Proficient with TensorFlow, PyTorch, Docker, Kubernetes, Cuda, Distributed Training, PyTorch Lightning, Git, Linux, OpenCV, CSS, React, AJAX, JQuery, Scikit-Learn, Pandas, Flask, SQLAlchemy, Nginx, MLOps, AWS, Sagemaker, Lambda, and other cloud-related technologies.
- Databases: MySQL, PostgreSQL, Cassandra
- Soft Skills: Strong in Analytical Thinking, Research, Problem-Solving, and Team Collaboration.
I am on a continuous quest for challenging opportunities in the fields of computer vision, machine learning, deep learning, and particularly in the deployment of these technologies in the cloud. My aspiration is to collaborate with a team that shares my enthusiasm for pushing the boundaries of AI and its practical applications.
Embark with me on this exciting journey, as we leverage the transformative power of AI and cloud technology to reshape our understanding and interaction with the digital world.
Experience
- Conducting advanced research in computer vision, specifically in image restoration using generative models like diffusion models, pix2pix. .
- Skills: Object Detection · Machine Learning Algorithms · Research · Applied Machine Learning · Deep Learning · Amazon Web Services (AWS) · PyTorch
- Spearheaded the research and development of advanced AI models for detecting and classifying brain-related diseases, including Large Vessel Occlusion (LVO), and hypodensity, with deep learning models like YOLOv7, detectron2 enhancing diagnostic accuracy and efficiency.
- Engineered and deployed robust APIs for the AI models, facilitating seamless integration into production environments. This deployment significantly improved the operational workflow and led to more reliable and rapid diagnosis of disease
- Pioneering Research in Synthetic Datasets: Spearheaded research initiatives on synthetic datasets derived from a limited number of authentic disease images. This innovative approach focused on enhancing data availability for robust model training, particularly in scenarios with scarce real-world data.
- Development of GANs for Disease Image Generation: Played a crucial role in designing and developing Generative Adversarial Networks (GANs). These models were adept at producing high-quality synthetic disease images. The use of these GAN-generated images was instrumental in significantly elevating the accuracy of disease identification models. This breakthrough demonstrated the potential of GANs in overcoming data limitations in medical imaging.
- Leadership and Training: As the team leader, led a group of researchers and developers in applying this cutting-edge research to product development. Additionally, undertook the responsibility of trainer, imparting essential skills and knowledge related to GANs and synthetic dataset utilization. This role encompassed mentoring the team to efficiently translate research findings into practical, market-ready healthcare solutions.
-
Spring 2022
- CACS354 - Advance Java Programming (3 cr)
- CACS355 - Network Programming with Java (3 cr)
-
Summer 2022
- CACS204 - Object Oriented Java Programming (3 cr)
- CACS402 - Cloud Computing (3 cr)
- CACS456 - Machine Learning (3 cr)
-
Spring 2023
- CACS354 - Advance Java Programming (3 cr)
- CACS355 - Network Programming with Java (3 cr)
- Work under the Prof. Dr. Suresh Manandhar
- Collaborated as a key member of an international team specializing in the research and design of machine learning and computer vision solutions for healthcare applications.
- Led the development of APIs and user interfaces for disease recognition models, enhancing the usability and accessibility of machine learning tools in medical diagnostics.
- Engineered a custom Machine Learning Pipeline tailored for medical image diagnosis, enabling dynamic, user-driven model training online for various diseases.
- Innovatively researched and developed a Deep Learning Medical Image Diagnosis Model, achieving significant performance improvements in detecting diabetic retinopathy, skin lesions, and lung diseases.
- Work under the Prof. Dr. Suresh Manandhar
- Developing language models to improve offline handwritten recognition tasks
- Synthesize handwritten images with GAN.
- Used of different types of GAN like Cycle GAN, Text2Image GAN, and Variational Encoder
- I Collaborated with a multidisciplinary product development team to identify performance improvement opportunities and integrate trained models
- Analyzed the data of customers, and performed customer segmentation, and built the predictive model to recommend products based on purchase history.
- I maintained, developed source code, and led a team of more than 3 people.
PROJECTS
Deep Learning Projects
Handwritten Line Text Recognition using Deep Learning with Tensorflow
- Tools: Flask, HTML, CSS, Bootstrap, Tensorflow, OpenCV
- This is my major project which I do with quiet research and found the state-of-the-art architecture for real-time handwritten recognition with CER 4.32%.
- Uses CNN, LSTM, CTC loss function and trained on an IAM dataset with a self-created dataset that can detect handwritten text in real-time.
- Perform data augmentation to make a robust model and improve accuracy.
- Also make some changes on the current architecture which is able to detect handwritten images in Devnagari Script with CER 8.32%.
Nepali Handwritten Line Text Recognition using Deep Learning with Tensorflow
- Tools: Flask, HTML, CSS, Bootstrap, Tensorflow, OpenCV
- This is my major project which I do with quiet research and found the state-of-the-art architecture for real-time handwritten recognition with CER 4.32%.
- Uses CNN, LSTM, CTC loss function and trained on an IAM dataset with a self-created dataset that can detect handwritten text in real-time.
- Perform data augmentation to make a robust model and improve accuracy.
- Also make some changes on the current architecture which is able to detect Nepali handwritten images in Devnagari Script with CER 10.32%.
A Nepali POEM Generation using AI
- Tools: Numpy, Tensorflow, keras, NLP
- This project use the char-rnn model.
- Train on Nepali POEM by Adikabi "Laxmi Prasad Devkota" who is pioneer in Nepali Literature with more than 200k characters.
- The char-RNN model is here used to generate the Nepali Poem
- Accuracy will be increase if more amount of data is available.
Reverse Search Engine with Transfer Learning and K nearest-neighbors.
- Tools: Numpy, Scikit-learn,Tensorflow v2, Flask
- Image Search Engine using Deep Learning Model (ResNet50)
- Use ResNet=50 model to extract features from Caltech101 datasets, train a K nearest-neighbors model using the brute-force algorithm to find the nearest n neighbors based on Euclidean distance.
- Based on less distance, top n images are returned.
- Also some analysis on how to increase speed and accuracy is given.
Other Projects
A simple and powerful Encryption and Decryption Application with own ASCII shifting algorithm.
Blog Posts & Paper Reading
In my free time, I usually create blog posts on fascinating deep learning and computer vision studies that go into extensive detail about ideas, architecture, and mathematical equations.
View all postsI annotated numerous interesting computer vision papers, particularly GANs and Vision Transformers, by providing an overview of each with details in mathematical calculations.
Read moreDeep Learning Paper Scratch Implementation with PyTorch & Tensorflow.
Read moreSkills
Languages and Databases
Libraries
Frameworks
Other
Certificates
Course Project Repository
Education
2023-2025| Starkville, USA
- Degree:Masters In Computer Science(Thesis Track)
- Concentration: Artificial Intelligence
- Relevant Coursework: Algorithms, Machine Learning, AI Robotics
- GPA: 4.0
2015-2019 | Kathmandu, Nepal
Degree: Bachelor In Computer Engineering
Grade: A
- For four years during my undergraduate studies, I received a merit-based scholarship for outstanding academic excellence.
- Completed Undergraduate In Computer Engineering with Distinction.
- Topper of my Batch in my University
- First position in instant software competition in my university.
- Data Structures and Algorithms
- Database Management Systems
- Operating Systems
- Artificial Intelligence
- Digital Image Processing And Pattern Recogntion
- Big Data Technologies
- Advance Mathematics
- Probability And Statistics
Accomplishments
Relevant Courseworks: