Selected work

Projects

Pipelines, automation, and a couple of things I built just to see if I could. Most of these run against real production data, warts and all.

Software Engineer — Mobile & API Automation, Deerhold Nepal

API & Mobile Automation Uplift

Took automated API test coverage from 20% to 95% and stood up a repeatable mobile QA workflow across Android and iOS using BrowserStack and Postman.

Inherited a service layer with almost no automated regression coverage and a manual mobile test cycle that ate days before every release. Designed API test scenarios directly against backend network traffic, then wired them into the sprint cycle alongside developers and product owners so regressions surfaced before code review instead of after release. In parallel, ran device-matrix mobile testing on BrowserStack across Android and iOS, cutting the manual verification pass needed before each build was signed off.

PostmanAppiumBrowserStackAPI Testing

Capstone Project, Institute of Science and Technology

Dominant Color Extraction with K-Means

Capstone project clustering image pixel data with K-Means to identify and rank an image's dominant colors.

Treats every pixel in an image as a point in RGB space and runs K-Means clustering to find the centroids that best represent the image's color palette. The resulting clusters are ranked by pixel share to produce a dominant-color breakdown, useful for tasks like palette generation, image tagging, and visual search.

PythonK-MeansMachine Learning

Personal Project

MeroJob IT Listings Scraper

A pet project scraping every IT job listing off MeroJob to study hiring trends across the Nepali tech market.

Pulled structured listing data — role, company, required skills, experience band, and posting date — off MeroJob's IT category pages, then cleaned and aggregated it with Pandas to surface which skills and seniority levels were most in demand at the time.

PythonBeautifulSoupPandas

Junior Software Developer, LIS Nepal

SQL Extraction Automation

Replaced an hours-long manual SQL-to-Python extraction routine with an automated script that finishes in minutes and self-reports completion via a Tkinter notification.

A senior engineer's manual process — hand-extracting SQL queries into a matching Python script, one file at a time — scaled linearly with the number of files and could burn most of a workday. Rebuilt it as a single automated pass that reads the source files, generates the corresponding query logic, and executes against the Snowflake warehouse directly, dropping total runtime to a few minutes and firing a Tkinter popup on completion.

PythonSQLSnowflakeTkinter

Software Engineer, Deerhold Nepal

Tiered Extraction Pipeline

A production data-acquisition pipeline that reaches for an API first, falls back to BeautifulSoup, and only spins up Selenium when a source is dynamically rendered or actively resists scraping.

Built around a simple rule: never do more work than the source demands. Each target site is profiled and routed through the cheapest extraction method that still returns complete data — API calls where available, BeautifulSoup for static HTML, and Selenium reserved for JavaScript-rendered or access-gated pages. Downstream, raw records pass through a normalization layer combining regex, XPath, and Fuzzy Wuzzy matching to reconcile inconsistent schemas, fix encoding issues, and de-duplicate near-identical entries before they reach analytics and reporting consumers.

PythonSeleniumBeautifulSoupXPathFuzzyWuzzy