projects
url-shortener
(TS, Fullstack)
websitecode
- • Implemented short-to-long URL redirection using middlewares, unit tests using Jest, data storage using Turso libSQL (SQLite)+Drizzle ORM, OAuth using Auth.js (NextAuth.js), server components using React, and Server Actions from Next.js.
rest-api-ts-sqlite
(TS, Backend)
websitecode- • Delivered a RESTful API to manage order and inventory data along with other technical requirements per the client's request.
- • Used Falso `@ngneat/falso` (a faster, tree-shakeable alternative to Faker.js) for testing purposes.
apollo-server-graphql-jwt-ts-simple-auth-backend
(TS, Backend)
code- • Built an auth backend that uses Apollo GraphQL and JWT (JSON Web Token) and is based on credentials (email and password). It is linked to a Turso libSQL (SQLite) database.
kampung-sengkemang
(HTML+CSS+JS, Frontend)
websitecode
- • Built an official website for a village hall following the client's request (the village head).
caffeineeee.me
(TS, Frontend)
code
- • Implemented CI/CD pipelines with `pre-commit` hooks and Biome.js.
- • Added gesture-based navigation for better UX.
- • Used AWS S3 for storing media files (images, etc.).
hacker-news-clone
(TS, Frontend)
websitecode
- • Built a fully responsive, 1-to-1 clone of the Hacker News website, https://news.ycombinator.com.
batak_toba_indonesian_nmt
(Python, Machine learning)
codeclick to view full image
- • Developed a Neural Machine Translation (NMT) model for translating between BBC (Batak Toba) and IND (Indonesian), a low-resource language pair, aiming to bridge the linguistic gap for Batak Toba speakers. Before June 2024, Batak Toba was not supported in Google Translate, making this model a pioneering effort in facilitating automated translation for the language.
- • Curated a parallel corpus by sourcing Batak Toba-Indonesian text pairs from various resources, including academic papers and community-driven translations.
- • Directly surveyed elderly, native Batak Toba speakers in Tebing Tinggi, North Sumatra to gather additional supporting data for linguistic analysis, validate translation accuracy, and include culturally relevant expressions often missing from written sources.
- • Preprocessed the data by tokenizing text, and fixed misaligned sentence pairs.
- • Achieved better results than all previous research in this area, proving that my approach works effectively and improves translation quality.
- • Try out the live models on HuggingFace, here:
- ϕ BBC -> IND: https://huggingface.co/kepinsam/bbc-to-ind-nmt-v5.
- ϕ IND -> BBC: https://huggingface.co/kepinsam/ind-to-bbc-nmt-v5.
- • Published a paper (10 pages) titled “Batak Toba language-Indonesian machine translation with transfer learning using NLLB (No Language Left Behind)” in IJAAS 2024 (indexed in Scopus Q4 and SINTA 1):
https://doi.org/10.11591/ijaas.v13.i4.pp830-839
.