Experience

AI for Scientific Knowledge Extraction – Developing intelligent agents that can parse and interpret scientific literature to extract polymer-related information, transforming unstructured PDFs into structured databases of polymer properties.

Phase Segmentation Techniques – Designing weakly supervised graph neural network models and unsupervised density-based methods to identify and segment phases in molecular dynamics data, enabling large-scale 3D analysis of atomic structures.

Active Learning for Molecular Data – Building active learning frameworks that intelligently sample from high-dimensional first-principles molecular dynamics datasets, ensuring diverse, information-rich training data for model development.

Machine Learning for Molecular Modeling – Applying deep and graph neural networks to predict atomic interactions, potential energies, and forces, accelerating molecular simulations and uncovering new insights into Earth’s interior and material behavior.

Grade and conduct lab for the course : COMP SCI II-MJRS [JAVA Programming] (CS 1351)

Grade and conduct office hours for the course : Applied Deep Learning (CSC4343).

Grade and conduct lab for the course : Introduction to Python (CSC2700).

Grade and conduct office hours for the course : Data Science & Analytics (CSC2730).

Grade and conduct office hours for the course : Interactive Computer Graphics (CSC4356).

Joined as one of the first 5 engineers and helped scale the platform and engineering organization to over 50 members, transforming Khalti from a growing startup into a nationwide fintech ecosystem.

Architected Khalti’s unified API middleware, integrating fragmented third-party services into a scalable, maintainable service layer that cut partner onboarding time from weeks to hours and streamlined long-term operations.

Developed a high-performance flight booking system that orchestrated complex API interactions, dynamic price aggregation, and caching strategies to optimize fares and reduce search latency across multiple vendors.

Built a fully automated financial reconciliation system for the accounts team, replacing manual, error-prone workflows. What previously took weeks or months to surface discrepancies became a daily automated reconciliation pipeline, drastically improving accuracy and financial transparency.

Enhanced platform reliability and fault tolerance by redesigning network and API communication layers, cutting vendor-side request failures by 40%.

Developed automated reporting and analytics workflows using Celery-based schedulers to deliver daily, weekly, and monthly transaction summaries directly to clients and finance teams.

Led DevOps modernization and release automation, ensuring consistent deployments across staging and production and enabling rapid iteration.

Mentored engineers and standardized documentation and review processes, strengthening team maturity, scalability, and delivery consistency.

Published a paper : Real Time Stock Market Prediction using Neural Network.

Build Real-Time Stock Data Analysis System.