Physics-Informed Deep Learning

Bisheshwor Neupane

My mission is to integrate Physics into Deep Learning Architectures, driving highly efficient and interpretable solutions for medical imaging and complex scientific problems.

View Research Focus LinkedIn
Profile Placeholder

Key Highlights:

  • • ARVO 2024 Publication
  • • ANAIS Hackathon Winner

01. Research Experience 🔬

My academic pursuit centers on fusing Physics directly into Deep Learning architectures, enhancing generalization, interpretability, and data efficiency for complex scientific problems.

3D Medical Imaging (JHU)

Pancreatic Tumor Classification

  • Developing advanced predictive models (CYST, PDAC, PNET).
  • Utilizing 3D CT scan analysis and radiomics feature extraction.
  • Focus on multi-modal data fusion for clinical decision support.
Computer Vision in Ophthalmology

Automated Optic Nerve Head Analysis

  • Developed a specialized CNN model for automated OCT-RNFL image analysis.
  • Quantified optic nerve head height (IIH patients).
  • Research presented at the prestigious ARVO 2024 conference.
Student Performance & Analytics (KU DLRL)

ML for Digital Learning Analytics

  • Led ML model training sandbox for student performance prediction.
  • Applied SVM for grade prediction and K-means for engagement analysis.

02. Industry Experience 💼

My professional experience focuses on architecting and optimizing production-grade AI systems, specializing in Large Language Models (LLMs) and Computer Vision for real-world enterprise needs.

AI Software Engineer (Ycotek)

Production-Grade Text-to-SQL Platform

  • Implemented sophisticated multi-tier memory systems (episodic, semantic, procedural) for context retention.
  • Engineered hybrid search techniques (fuzzy, semantic, partial matching) for superior query retrieval.
  • Optimized platform for low-latency SQL generation.
AI Software Engineer (Ojas)

AI Ad Generation Platform

  • Developed LLM-based agent systems for automated ad generation.
  • Designed and implemented VLM pipelines for ad replication.
  • Built a robust document extraction platform.
Machine Learning Engineer (Fusemachines)

Vision-Language & Conversational AI

  • Built an LLM-based chatbot integrating Vision-Language Models.
  • Contributed to an AI interviewing agent using probabilistic head motion synthesis.
  • Optimized CNN and GAN models using advanced hyperparameter tuning.

03. Education & Milestones 🎓

Education

Bachelor of Computer Engineering

Kathmandu University (2018-2023)

CGPA: 3.7/4.0

Key Achievements

  • Winner: ANAIS 2024 Startup Challenge Hackathon (Multi-agent Proposal Generation System)

  • Completed intensive Nepal AI School - NAAMII (Geometric Deep Learning, 3D Computer Vision, Trustworthy AI).

  • Published in ARVO 2024 for automated ophthalmology analysis.

04. Insights & Blog 💡

A space for exploring deeper technical concepts, new research directions, and my thoughts on the future of Data Science, LLMs, and PI-DL.

The Case for Physics-Informed Neural Networks

October 15, 2025 | PI-DL

A deep dive into how incorporating physical constraints dramatically improves model stability, reduces data requirements, and enhances interpretability.

Read Article (Placeholder)

Architecting Multi-Tier Memory for LLMs

September 28, 2025 | LLM Systems

Explaining the implementation of episodic, semantic, and procedural memory systems for conversational agents beyond short-term context windows.

Read Article (Placeholder)

3D Feature Extraction in Medical Diagnostics

August 20, 2025 | Medical AI

Analyzing the challenges and advances in using radiomics and volumetric analysis to enhance tumor classification from raw CT data.

Read Article (Placeholder)

Let's Connect 👋

I am actively seeking collaborations related to Physics-Informed Deep Learning, and 3D medical diagnostics. Reach out to discuss research or projects!

Email Me Directly