- Implemented and optimized an end-to-end AI system on AWS, reducing inference latency and model training costs, resulting in significant time and cost efficiency improvements.- Created a multimodal AI model integrating sensor data to provide accurate health metrics, enhancing user health management through advanced data fusion and attention mechanisms.- Adapted and fine-tuned an open-source LLM model for behavior recommendations, improving the relevance and… Show more - Implemented and optimized an end-to-end AI system on AWS, reducing inference latency and model training costs, resulting in significant time and cost efficiency improvements.- Created a multimodal AI model integrating sensor data to provide accurate health metrics, enhancing user health management through advanced data fusion and attention mechanisms.- Adapted and fine-tuned an open-source LLM model for behavior recommendations, improving the relevance and effectiveness of user guidance.- Enhanced personalization of behavior recommendations by integrating Retrieval-Augmented Generation (RAG) into the fine-tuned LLM model, delivering more accurate and contextually relevant advice.- Optimized multimodal and LLM models using RLHF based on user interactions, significantly enhancing user satisfaction and engagement.- Developed a multi-agent system with LangChain, improving the robustness and performance of the AI system through better-coordinated agent interactions.- Facilitated cross-functional collaboration, incorporating feedback from multidisciplinary teams via Jira and Confluence, resulting in successful launches of new app features.- Conducted extensive user testing, including A/B testing, refining algorithms and user interfaces with Jupyter Notebooks and UserTesting.com to deliver a user-centric app experience that received high satisfaction from early adopters.- Managed and processed large datasets using tools and technologies like Apache Spark, Hadoop, and AWS S3, ensuring data quality through techniques such as data cleansing, normalization, and validation, which improved model accuracy and reliability.- Actively working on distributed model training, model quantization, small language models, and LLM routing to enhance deployment scalability and practicality.- Led cross-team collaboration efforts and effectively managed an offshore development team resulting in enhanced team productivity. Show less

- Integrated and customized Wav2Vec 2.0 model to improve accuracy in STT for supporting users with mental health conditions, leading to more reliable speech recognition.- Enhanced user engagement and communication by implementing advanced TTS technology using Google Cloud Text-to-Speech API, Tacotron 2, and FastSpeech models.- Minimized inference latency and enhanced training efficiency by conducting kernel fusion optimization and implementing model optimization… Show more - Integrated and customized Wav2Vec 2.0 model to improve accuracy in STT for supporting users with mental health conditions, leading to more reliable speech recognition.- Enhanced user engagement and communication by implementing advanced TTS technology using Google Cloud Text-to-Speech API, Tacotron 2, and FastSpeech models.- Minimized inference latency and enhanced training efficiency by conducting kernel fusion optimization and implementing model optimization techniques (e.g., model compression and knowledge distillation) and inference techniques (e.g., frame skipping and look-ahead mechanisms) using tools like TensorRT and cuDNN.- Managed and processed extensive voice datasets and associated metadata using tools like Google Cloud Storage and Google Cloud Dataflow, implementing data preprocessing techniques such as noise reduction, normalization, and feature extraction to ensure high-quality inputs for STT and TTS models, resulting in improved model performance and reliability. Show less

• Designed and conducted 12+ interactive workshops to translate complex ML concepts from over 30 research papers into actionable insights for non-technical audiences, increasing the use of advanced analytic techniques in clinical research and enhancing diagnostic accuracy.• Developed the first foundational ML model trained on extensive ECG data using self-supervised learning and GANs, extracting detailed heart-rate metrics and other cardiovascular signals to improve clinical diagnostics and… Show more • Designed and conducted 12+ interactive workshops to translate complex ML concepts from over 30 research papers into actionable insights for non-technical audiences, increasing the use of advanced analytic techniques in clinical research and enhancing diagnostic accuracy.• Developed the first foundational ML model trained on extensive ECG data using self-supervised learning and GANs, extracting detailed heart-rate metrics and other cardiovascular signals to improve clinical diagnostics and patient monitoring.• Led the design and development of large-scale distributed web crawlers using Microsoft Azure Functions and Power Automate, and maintained scrapers with Microsoft Playwright for dynamic content, enhancing data collection for clinical research.• Created and managed data acquisition pipelines using Azure Data Factory and Event Hubs, optimizing performance with asynchronous processing on Microsoft Azure, and developed dashboards with Power BI to enable clinical staff to visualize and utilize web crawler output effectively. Show less

• Led the enhancement of an XGBoost model using EHR and pediatric ICU data, achieving a 15% improvement in early prediction of Acute Kidney Infections (AKI) and enabling critical interventions 72 hours before onset.• Developed key features for AKI management, directly addressing business and patient care needs for improved patient monitoring and aligning with business goals to capitalize on features unique to company PICU devices.• Contributed to preparing a patent application for the… Show more • Led the enhancement of an XGBoost model using EHR and pediatric ICU data, achieving a 15% improvement in early prediction of Acute Kidney Infections (AKI) and enabling critical interventions 72 hours before onset.• Developed key features for AKI management, directly addressing business and patient care needs for improved patient monitoring and aligning with business goals to capitalize on features unique to company PICU devices.• Contributed to preparing a patent application for the novel predictive model, showcasing the ability to engage in high-level intellectual property initiatives.• Successfully published a journal paper based on the research and presented at industry conferences, enhancing the company’s reputation as an innovator in health tech and generating interest from potential partners. Show less

• Designed the pilot Nepal Public Service Fellowship program under the supervision of the Chief Advisor to the Prime Minister, creating a structured initiative to address high-priority and performance management project needs.• Performed a needs assessment of short-term and long-term projects through in-person interviews and online surveys with potential hosts and fellows, identifying key project requirements and aligning the fellowship design with organizational goals.• Conducted a… Show more • Designed the pilot Nepal Public Service Fellowship program under the supervision of the Chief Advisor to the Prime Minister, creating a structured initiative to address high-priority and performance management project needs.• Performed a needs assessment of short-term and long-term projects through in-person interviews and online surveys with potential hosts and fellows, identifying key project requirements and aligning the fellowship design with organizational goals.• Conducted a participatory, iterative refinement process based on feedback from potential fellows and hosts, enhancing the fellowship design’s relevance and effectiveness.• Recommended the final fellowship design to senior government officials, securing approval and contributing to the strategic development of public service initiatives in Nepal. Show less

• Led and mentored an interdisciplinary team to successfully meet project milestones, demonstrating strong leadership and effective team management using Jira for project tracking.• Developed and implemented advanced analytical tools for clinical research, including predictive analytics and anomaly detection models using logs and disparate data sources, leveraging Python, R, and scikit-learn.• Optimized data processing pipelines with Apache Hadoop and Apache Spark to enhance time… Show more • Led and mentored an interdisciplinary team to successfully meet project milestones, demonstrating strong leadership and effective team management using Jira for project tracking.• Developed and implemented advanced analytical tools for clinical research, including predictive analytics and anomaly detection models using logs and disparate data sources, leveraging Python, R, and scikit-learn.• Optimized data processing pipelines with Apache Hadoop and Apache Spark to enhance time efficiency and facilitate anomaly and outage detection in healthcare data systems.• Applied data analysis and metrics with SQL and Tableau to continuously improve software systems and processes, incorporating user feedback to refine functionalities.• Followed the DevOps model in developing and deploying data processing solutions using Jenkins for CI/CD, ensuring robust, scalable, and reliable operations.• Fostered innovation through collaboration and mentorship, guiding junior data scientists and contributing to high-impact research publications. Show less

• Orchestrated backend development and web interface design, leveraging Apache Hadoop and Hive to improve data processing (MRI, PET) efficiency by 40% and enhance diagnostic accuracy for clinicians, significantly streamlining operations and supporting more accurate clinical assessments.• Implemented RESTful APIs using Django, optimizing medical image processing and bolstering research capabilities. Integrated with Apache Hadoop for handling large-scale data processing, which was pivotal in… Show more • Orchestrated backend development and web interface design, leveraging Apache Hadoop and Hive to improve data processing (MRI, PET) efficiency by 40% and enhance diagnostic accuracy for clinicians, significantly streamlining operations and supporting more accurate clinical assessments.• Implemented RESTful APIs using Django, optimizing medical image processing and bolstering research capabilities. Integrated with Apache Hadoop for handling large-scale data processing, which was pivotal in securing seven publications and three grants, underscoring its impact on the institute's research output.• Enhanced the accuracy of machine learning models for medical data analysis by developing and fine-tuning models using scikit-learn to analyze MRI and PET scans. Utilized Hadoop MapReduce for advanced feature engineering and large-scale data transformations, resulting in a 20% increase in model accuracy and more reliable diagnostic support for clinicians.• Ensured the robustness and scalability of data processing systems by conducting performance optimization and scalability testing for backend systems with Apache Hadoop, achieving a 50% reduction in data processing time and improved system reliability. Show less