About

Mobile engineer since 2022.Kathmandu, working on US time.

Mobile engineer since 2022, Flutter on top and native Kotlin underneath. Seven products across thirteen Google Play listings, one past 100,000 installs.

Research›Full CV›

12h

the gap between Kathmandu and the US clients I lead mobile delivery for, inside a twenty person team. Client meetings and demos happen on video, in English.

In short

I lead mobile delivery for US clients out of Kathmandu, inside a twenty person team, across a twelve hour gap. Client meetings and demos happen on video, in English.

I own the parts teams usually outsource. A forked native video player, subscription billing and the release pipeline itself. Before that, a national internet provider, where a bad release once stopped the support line. That is where the release discipline comes from.

Research

Whether code produced by a language model respects the lifecycle and threading contracts of the platform it runs on. A clean compile does not prove it. Neither does a passing test, because the test rarely puts the code in the state where the contract matters. Four years of writing that code by hand is where the question came from, and it is the direction I want for graduate work.

Low latency audio and video playback on hardware with no performance headroom to spare.

Offline first synchronisation, where the device rather than a server is the unit of correctness, which turns conflict resolution into the hard part.

RAGE, real time age, gender and emotion detection

Python, TensorFlow, MTCNN, OpenCV

BE final year project, team of four, 2021 to 2022

MTCNN aligns the face first. Three CNN heads then read the same cropped frame in parallel, so one webcam stream returns all three predictions at once. Alignment is what makes it work. It holds the predictions steady under poor light and a tilted head.
Validation accuracy 79.0% on emotion (FER-2013, 16,175 images), 94.4% on gender (2,307 images), 62.9% on age (23,708 images). Live inference ran at roughly 4 to 5 frames per second on a laptop webcam.
The report states its own limits. Age predictions pull toward the peak of the training distribution, the gender model skews male because the dataset does, and accuracy was capped by dataset size and camera quality.

Supervised by Prof. Dr. Subarna Shakya, IOE Pulchowk Campus, Tribhuvan University.

Education

Bachelor of Engineering, Computer Engineering

Kantipur Engineering College, Tribhuvan University, Lalitpur, Nepal

2017 to 2022

Tribhuvan University First Division. Final year project RAGE, supervised by Prof. Dr. Subarna Shakya, IOE Pulchowk Campus.

Languages

Nepali: Native
English: C1, IELTS 7.0
German: Basic

IELTS Academic, overall 7.0, CEFR C1.

Full CV›