Pawan Shrestha

Built a U-Net model predicting urban sprawl across 3 Nepali cities at 85% accuracy.

Recent Computer Engineering graduate with a strong foundation in full-stack web development, machine learning, and AI systems.

View Projects Resume
◈unet_model.py
1def build_unet(input_shape=(256, 256, 3)):
2 # Urban Sprawl Prediction - U-Net architecture
3 inputs = Input(shape=input_shape)
4 # Encoder path
5 c1 = Conv2D(64, 3, activation='relu')(inputs)
6 p1 = MaxPooling2D()(c1)
7 c2 = Conv2D(128, 3, activation='relu')(p1)
8 return Model(inputs, outputs)
9 # → 85% accuracy on satellite classification

About Me

I build machine learning models and turn them into full-stack applications.

Pawan Shrestha portrait
Bhaktapur, Nepal
Open to junior software, AI, and data science opportunities

I'm Pawan Shrestha, a Computer Engineering graduate from Nepal Engineering College. I spent my final year training a U-Net on satellite imagery to forecast urban sprawl across Kathmandu, Pokhara, and Hyderabad, it hit 85% accuracy on land-classification. I also build practical web apps end to end: a complaint-reporting platform for city maintenance and a music-sharing social app. I write Python, and I care about models that hold up outside a Jupyter notebook: real data pipelines, real APIs, real deployments. I'm looking for a junior role in software engineering, ML, or data science where I can do that work for a product with users.

BuiltU-Net for urban sprawl — 85% accuracy
LanguagesPython, TypeScript
Pipelinepandas → TensorFlow → deployed API

Skills

Tools I reach for when the work is real.

Databases
MongoDBSQL
Languages
JavaScriptPythonHTMLCSS
Frameworks & Libraries
ReactPandasNumPy
Developer Tools
VS CodeGitGitHubJupyter Notebook
Concepts
Machine LearningData ScienceREST APIs

Projects

Some of my projects

Screenshot of Urban Sprawl Prediction

Urban Sprawl Prediction

A U-Net that forecasts where Kathmandu, Lalitpur, and Bhaktapur will grow by 2030.

Trained a U-Net on Landsat satellite imagery to classify built-up vs non-built-up land, reaching 85% accuracy across the three cities. I chained it into a ConvLSTM that extends those classifications forward, so the model outputs 2030 sprawl maps at 78% precision. The pipeline — image tiling, augmentation, training, and map rendering — is all pandas, scikit-learn, and TensorFlow, and I wrote the visualizations that turn predictions into readable growth maps.

PythonScikit-learnPandasTensorflow
Screenshot of AI Agent with RAG & Tool-Calling

AI Agent with RAG & Tool-Calling

A Python AI agent built with LangChain. It answers questions from general knowledge, from your own local documents (RAG), from live web search, or with a calculator

An LLM-powered conversational agent that combines Retrieval-Augmented Generation (RAG) with dynamic tool-calling and memory. Built using LangChain and the Gemini API, it ingests custom documents into local vector databases (Chroma) for grounded context and executes live web searches for real-time query routing, all accessible via an interactive Streamlit UI.

PythonPythonLangChainGemini APIChromaStreamlit
Screenshot of CleanCity

CleanCity

A civic reporting platform where a photo and GPS pin log a city issue for officials to resolve.

Residents upload a photo of a problem — garbage, a pothole, a blocked drain — with their location. I built the classification step with an AI model that identifies the issue type so it routes to the right department, then an admin dashboard where officials track and close each complaint. It runs on Next.js with a Supabase backend and OpenStreetMap for geolocation.

Next.jsReactTypeScriptSupabaseOpenStreetMap
Screenshot of Near Beat

Near Beat

A full-stack web application that helps people discover, join, and chat about nearby real-world events in real time.

NearBeat is a full-stack event discovery platform that connects users with nearby local activities. It features location-based event search, real-time participant chat, instant notifications, and secure user authentication built on a robust React, Node.js, and PostgreSQL stack.

ReactTypeScriptNode.jsExpress.jsPostgreSQL
Screenshot of Melomatch

Melomatch

A social app where strangers post a song and discover what others are listening to.

Users create profiles, post the song they can't stop replaying, and browse strangers' picks. I built the REST API in Node/Express with MongoDB for profiles and posts, and the React frontend handles auth, feed pagination, and a like system. The interesting part was designing the data model so a post can surface three different feed orders — recent, liked, and from people with overlapping taste.

ReactNodejsMongoDB

Experience

Where I have worked

May, 2025 — Oct, 2025

Web Development Intern

Aramex · Kathmandu, Nepal

As a Frontend Developer Intern at Aramex, I resolved React component rendering issues, fixed functional UI bugs, and improved overall application stability under the mentorship of senior engineers. I also maintained client-facing features in the customer portal, including updating delivery service locations to support expanding logistics operations.

Education

Four years of engineering, one thesis.

+2 ScienceJul, 2019 — Apr, 2021

Khwopa College

Bhaktapur, Nepal

Physics, chemistry, and math, the coursework that pushed me toward engineering and eventually programming.

Bachelor's in Computer EngineeringFeb, 2022 — Mar, 2026

Nepal Engineering College

Bhaktapur, Nepal

Final-year thesis was the urban sprawl U-Net. Core coursework covered algorithms, databases, OS, networks, and AI and I pulled most of it into hands-on projects rather than leaving it in lectures.

Certifications

Courses that changed how I build.

Cohere Labs
2026

ML Summer School 2026

Participated in an applied machine learning summer school program, engaging with current ML research and collaborative learning.

Verify on Cohere Labs
Oracle
2026

Agentic AI Foundations Associate

Certified foundational knowledge of agentic AI systems and automation.

Verify on Oracle
DataCamp
2024

Intermediate Python

Intermediate Python: functions, data structures, and the pandas/NumPy workflow I now use daily for data analysis.

Verify on DataCamp
Microsoft and LinkedIn
2026

Career Essentials in Generative AI

Explored core concepts of artificial intelligence, ethical considerations, prompt engineering techniques, and practical strategies for utilizing generative AI.

Verify on Microsoft and LinkedIn

Contact

Let’s build something together.

Have a project, a role, or an idea you’d like to talk about? I’d love to hear from you.

pawanshrestha0702@gmail.com
Bhaktapur, Nepal
Email me directly