Portfolio — 2026
Kathmandu → Brussels
Frontend engineer · React & Next.js
I build the part people actually touch. Nine production surfaces so far — marketing sites, multi-tenant apps, AI dashboards — and on most of them I was the only frontend engineer on the team.
Seven years in: started by co-founding a web agency in eastern Nepal, spent a stretch in backend and applied machine learning, then came back to the frontend for good. Next stop is Brussels, for an MSc at VUB.
Index card
Chapter 01
2019 — 2023
Birtamode, Nepal
LinkBinary · Co-founder
We started LinkBinary out of Birtamode. I did the technical side: scoping calls, architecture, the actual build, and the demo where a client asks why a button is that colour.
Eleven-plus delivered projects later, the lesson that stuck wasn't a framework. It was that a site only counts once it is live, loading fast, and someone who isn't a developer can run it.
Selected client work · shipped as part of the team
Marketing site for an IB candidate school in Jhapa — programme pages, admissions, heavy content structure.
takshashilaschool.edu.np
Savings-and-credit cooperative site: member information, services, trust-first layout for a finance audience.
aabaddha.coop.np
Restaurant presence built around ordering and hall reservation — menu, gallery, conversion paths.
pizzaghar.com.np
The agency itself — the company site we used to win the work above.
linkbinary.com.np
Chapter 02
2025
Kathmandu · hybrid
ChimpVine · AI intern → trainee engineer
Seven months of backend and applied AI: FastAPI and Django services, AWS architectures, computer-vision experiences, and automation pipelines that removed work nobody enjoyed doing.
It made me a better frontend engineer, mostly by making me honest about where complexity belongs. I also kept research running on the side — two papers are under review at Springer Nature.
Services
FastAPI, Django and Node services with auth, rate limiting and secured REST endpoints.
Cloud
AWS Lambda, S3, CloudFront, EC2 and RDS, wired to automated deployments.
Applied AI
LLM features for content and workflow, OpenCV interactions, n8n automation pipelines.
Chapter 03
2025 — now
Remote · Austin & beyond
OpsAway · Lead frontend
Ninjahire · Frontend
On most of these products there is no second frontend engineer to hand things to. I set the architecture, build the component system, ship the marketing site and the app behind it, and keep it stable while the product changes underneath me.
Two of them are large team products, where the job is fitting into someone else's patterns without breaking production. Both modes are on the table below.
Production surfaces
01 / 09
Chapter 04
Sep 2026
Brussels / Leuven
VUB · MSc Computer Science
From 3 September 2026 I am in Belgium for a Master of Applied Sciences and Engineering: Computer Science at Vrije Universiteit Brussel. I want to keep shipping while I study.
Chapter 05
Product · TypeScript
Bookkeeping interface: ledgers, entries and totals that stay readable as the data grows.
Live demo
Product · React
Application frontend built alongside the Opolis landing page.
Live demo
Product · TypeScript
Marketing and product surfaces rebuilt around a single component set.
Live demo
Product · React
Client-facing frontend with a typed API layer and reusable layout primitives.
Live demo
Applied AI
DenseNet classifier and U-Net segmentation on breast ultrasound, served through Django with a React frontend.
Repository
Applied AI
Retrieval-augmented question answering over a document set.
Repository
Applied AI
Eleven architectures implemented from scratch in PyTorch, across vision, language and generative models.
Repository
Applied AI
Real-time detection and extraction with YOLOv8 and OpenCV.
Repository
Chapter 06
Research
Two papers · under review
Springer Nature
Medical imaging · deep learning
DenseNet121, EfficientNet-B1 and ResNet50 compared under transfer learning, with Mixup and label smoothing. EfficientNet-B1 held the best metric profile at the lowest computational cost.
Status — under review
Bioinformatics · machine learning
A multi-model machine learning approach
Random Forest and XGBoost over transcriptomic data surfaced GUCA2A, SCARB2 and ZNF91 as candidate diagnostic markers.
Status — under review
Interface
Services
Infrastructure
Applied AI
Recognition
Certification
Outside the editor
Contact
Send me the product and the problem. I will reply with what I would build first, and links you can open without a login.