Hi, I’m Julesh. I build things end to end.
I move across the whole stack — backend, frontend, infrastructure, and every so often a neural net I built from scratch. Lately that’s meant building the reporting and payment systems behind Dimitra’s agricultural supply chain, freelancing for a trading-card marketplace on the side, and putting my Master’s in AI to more use than I expected.
About
Things I’ve shipped
7 projectsAmong everything I’ve built, these are the ones that mattered most — or taught me the most getting there.
Due-Diligence & Deforestation Reporting
A due-diligence API that runs a deforestation check, generates the PDF a regulator reads, and records the result on Solana — authenticated with per-user secret keys, with the chain write kept async so a slow node can't hold up compliance reporting.
Farmer ID Card & Online Verification
Server-side PDF, QR and vCard generation, an admin issuance screen, and a public page anyone can scan to verify a farmer's identity — kept in sync across three codebases.
Carbon Credit Agreements
Replaced manual paperwork with signed PDF agreements, presigned S3 uploads, and full assignee and payment history for carbon-credit projects.
PSA Trading-Card Data Pipeline
Freelance work for HobbyX Marketplace — an async ETL pipeline that fetches trading-card and grading data from PSA and loads it into MySQL as a scheduled job.
HobbyX — Grading Tracker App
Co-built the mobile app for HobbyX's grading-tracker service — shared backend and app-side frontend work, and owned the DevOps pipeline through to App Store and Play Store release.
PSA Card Recognition Tool
An internal tool for HobbyX that turns a card photo into a ready-to-submit PSA search query — computer-vision matching against a known-card catalog, with an AI fallback for anything it doesn't recognize confidently.
Wilms Tumor Segmentation
MTech thesis — compared UNet, DeepLabv3 and SegFormer for segmenting pediatric kidney tumors across 10,000+ CT scans, reaching up to 99.60% pixel accuracy and a 97.18% Dice score.
Where I’ve worked
4 rolesAn Express/Sequelize monolith, a Vue admin panel, a NestJS GraphQL subgraph, and a public verification SPA — four repos in service of one agri-supply-chain platform.
Independent client work spanning data pipelines to full-stack products — most notably for HobbyX Marketplace, an online trading-card marketplace and grading service.
↗ hobbyxmarketplace.comInternal tools and open-source-adjacent fixes for a small dev shop.
Where it started — HTML, CSS, JS, Node, and a first pass at MongoDB.
What I work with
Kathmandu University
2 degreesKathmandu University · Panchkhal, Kavre
MTech, Artificial Intelligence
Thesis: Wilms Tumor Segmentation using Deep Neural Networks. Compared UNet, DeepLabv3 and SegFormer across 10,000+ CT images from the TCIA dataset, reaching up to 99.60% pixel accuracy, 94.51% IoU and a 97.18% Dice score.
Kathmandu University · Dhulikhel, Kavre
BSc, Computer Science
Final project: Age and Gender Classification from Facial Images using VGG, focused on handling limited datasets. Achieved 96.29% gender-prediction accuracy on WIKI dataset images.
Let’s build something.
Open to full-stack roles and interesting side projects. Based in Kathmandu, Nepal, working remote with teams anywhere.
I’m a full-stack developer who moves between a NestJS API, a Vue admin panel, and a stray PyTorch model in the same week. Most of my time goes into Dimitra, building the systems that keep an agricultural supply chain traceable and compliant. Alongside that: freelance work for HobbyX, a trading-card marketplace — a data pipeline, a mobile app, and the computer-vision tooling behind it — plus a Master’s in AI that started with segmenting tumors in CT scans before it turned into a career shipping code.