Skip to content

$ whoami

Bhagwan Prasai

Final-year Computer Engineering student with a published research paper, a shipped full-stack ML application, and active research in medical AI. Building deep learning pipelines and cloud-native services, one project at a time.

See my work Read my paper
0
Published paper
0
Active projects

Currently: studying for AWS Cloud Practitioner · Open to internships (Nepal & remote)

About

Final-year Computer Engineering student at Kantipur Engineering College. I work close to the fundamentals — training deep learning models, building cloud-native backends, and reading the papers instead of stopping at the framework. Published a hybrid CRNN + Seq2Seq attention model for handwriting recognition; currently extending that into LungARC, a 3D lung nodule analysis pipeline. Also building CloudStudy in ASP.NET Core with LocalStack, studying for AWS Cloud Practitioner, and open to internships in ML engineering, cloud-native, or backend — Nepal & remote.

Skills

Languages

  • Python
  • C#
  • C
  • C++
  • JavaScript basic
  • HTML / CSS

ML & Research

  • PyTorch
  • OpenCV
  • NumPy
  • scikit-learn
  • MONAI
  • MediaPipe MediaPipe

Backend & Cloud

  • ASP.NET Core
  • FastAPI
  • Docker basic
  • AWS Cloud Practitioner — studying
  • LocalStack LocalStack

Tools

  • Git & GitHub
  • VS Code
  • Linux
  • Jupyter

Projects

01

Air Writing Recognition

HTR Web App

Completed

Python · FastAPI · PyTorch · MediaPipe · OpenCV · WebSocket · SQLite · uvicorn · Pydantic

Real-time gesture-based text input — draw characters in the air via webcam and the app turns hand movements into digital notes. Reuses the CRNN + Bahdanau attention architecture from my paper, adapted for live inference.

Source ↗
02

LungARC

3D Lung Nodule Analysis · Final-year major project

Active research

Python · PyTorch · MONAI · SimpleITK · pylidc · NumPy · scikit-image

A deep learning pipeline for automated lung nodule detection, 3D volumetric reconstruction, and longitudinal growth tracking from CT scans — reasoning across time, not just single scans. Final-year major project, team of four.

03

CloudStudy

Cloud-Native Productivity Suite

Early development

C# · ASP.NET Core · LocalStack (S3, SQS, Lambda) · Microservices · REST APIs

A microservices-based productivity platform for students, built locally with LocalStack emulating AWS — practical cloud-native architecture (S3, SQS, Lambda) without the cloud bill.

04

AirWriting — AWS Deployment

Learning to deploy an ML app to the cloud, one service at a time

In progress

Docker · AWS ECR / ECS / Fargate · IAM · CloudWatch · ALB · Floci (local AWS emulator)

Walking the AirWriting ML app through real AWS deployment one service at a time — ECR, ECS Fargate, IAM, CloudWatch, ALB — using Floci as a local AWS emulator so I can break things without paying for a real account. App is currently running under ECS Fargate; working through the ALB.

Source ↗ Related app ↗

Publication

KEC Conference 2026 · The Special Issue of InJET

Hybrid CRNN with Seq2Seq Attention Mechanism for Handwriting Recognition

Bhagwan Prasai, Avishek Gautam, Shruti Sapkota, Biniv Maharjan, Nirajan Acharya

ISSN: 3021-940X (Print) · DOI: 10.3126/injet.v3i2.95148 ↗

  • Key result — 13.59% WER and 4.28% CER on the IAM handwriting dataset (115,320 word images, 657 writers) — outperforming Kass et al. (2022) by 1.81% WER and 2.22% CER, and Shrestha et al. (2023) by 5.05% CER.
  • Architecture — 10-layer VGG-style CNN with residual connections + DropPath → BiLSTM encoder → Bahdanau attention + LSTM decoder (seq2seq).
  • Novel contributions — DropPath inside CNN residual blocks (first in attention-based HTR encoders), combined EMA + SWA + adaptive teacher forcing decay without synthetic data, and dynamic attention masking for variable-width batches.
  • Validated — Attention heatmaps confirm diagonal spatial-sequential alignment, proving the model decodes characters left-to-right rather than randomly attending to image regions.

Education

2023 — 2027 (Expected)

Bachelor of Engineering — Computer Engineering

Kantipur Engineering College, Tribhuvan University · Dhapakhel, Lalitpur, Nepal

Relevant coursework: Data Structures & Algorithms, Artificial Intelligence, Computer Networks, Operating Systems, Software Engineering, Database Management Systems.

Certifications
  • Machine Learning Certification Kantipur Engineering College
  • AWS Certified Cloud Practitioner in progress Amazon Web Services
  • Cybersecurity Fundamentals in progress Online
Languages

Nepali (native) · English (professional) · Hindi (conversational)

Contact

Open to internship and entry-level roles in ML engineering, cloud-native, or backend development — Nepal & remote. Feel free to reach out.