Saurav Shrestha

SauravShrestha

Computer Engineering student at Kathmandu Engineering College, turning ideas into things that solve real-world problems.

View projectsDownload CV
Scroll

About

Curious by default

I’m an undergrad Computer Engineering student at Kathmandu Engineering College, driven by curiosity and a passion for building things that matter. My interests and expertise lie in Machine Learning, Deep Learning, and Data Science, where I enjoy turning ideas and data into meaningful solutions.

Beyond what I already know, I’m always eager to explore the unexplored. I love diving into new fields, learning new technologies, experimenting with unconventional ideas, and challenging myself to understand how things work. For me, every new concept is an opportunity to create something different.

I’m constantly learning, building, experimenting, and looking for the next problem worth solving.

  1. Kathmandu Engineering College

    B.E. Computer Engineering · Kathmandu, Nepal

     — 

  2. Kathmandu Model College

    NEB Class XII — Science · Kathmandu, Nepal

     — 

  3. Sainik Awasiya Mahavidyalaya

    Secondary Education Examination (SEE) · Nepal

     — 

Selected work

Projects

Computer Vision

Ensemble of 3

AI Image Detector

Web app that classifies an uploaded image as AI-generated or a real photograph, combining two independent vision models into a stacked ensemble verdict.

  • Python
  • TensorFlow
  • PyTorch
  • MobileNetV2
  • Vision Transformer
  • Flask
View detailsHide details
  • Trained a MobileNetV2 convolutional network and a Vision Transformer, then stacked both through a logistic-regression meta-model.
  • Surfaced texture, lighting, edge and colour measurements beside every verdict so the result can be questioned rather than trusted blindly.
  • Built with a team of four and served through a Flask API with a browser front end.

Predictive Modelling

2,211 Listings

Nepal Housing Price Model

Regression pipeline predicting asking prices for Kathmandu-valley property, built around a parser for Nepali land units and dates.

  • Python
  • pandas
  • scikit-learn
  • Random Forest
  • Feature engineering
View detailsHide details
  • Converted Ropani–Aana–Paisa–Dam areas onto a single scale and Bikram Sambat build years to Gregorian.
  • One-hot encoded 33 amenities, taking 18 raw columns to 74 usable features.
  • A random forest reached R² 0.53 — a rough price band, not a valuation for a specific house.

Neural Networks

98% Recall

Early Breast Cancer Detection

A classifier that separates malignant from benign breast tumours on scikit-learn’s Wisconsin diagnostic dataset — a single-layer neural network written by hand in NumPy, with the decision threshold tuned so a missed cancer costs more than a false alarm.

  • Python
  • NumPy
  • scikit-learn
  • From scratch
  • Gradient descent
View detailsHide details
  • 569 biopsies with 30 measurements each, split 75/25 with stratification and standardised before training.
  • The network is raw NumPy — sigmoid forward pass, error derivative and weight updates across 5,000 epochs, no framework doing the learning.
  • Pushing the threshold to 0.9 catches 52 of 53 malignant cases (98% recall) at 95.1% overall accuracy and an ROC AUC of 0.996 — it trades 6 false alarms for 1 missed cancer, which is the right way round.
  • A benchmark dataset and a learning exercise, not a diagnostic tool.

Toolkit

What I work with

Technical Workflows

  • Python
  • C
  • C++
  • JavaScript
  • Swift
  • SQL
  • HTML
  • CSS
  • TensorFlow
  • Keras
  • PyTorch
  • scikit-learn
  • NumPy
  • pandas
  • Flask
  • Django
  • Flutter
  • Git
  • GitHub
  • Jupyter
  • Google Colab

Creative Workflows

  • DaVinci Resolve
  • CapCut
  • Lightroom
  • Canva
  • Photoshop
  • Audacity

Let’s connect

Get in touch

Open to internships, collaborations, and anything that sounds like a good problem.

LinkedInGitHubEmail

sauravstha1080@gmail.com
Kathmandu, Nepal

Curriculum vitae

Download PDFUpdated September 2026