About Digistar Class
Digistar Class is a national digital talent acceleration program organized by PT Telkom Indonesia to train software engineers, product managers, and UI/UX designers across three specialized tracks: Hacker, Hipster, and Hustler.
Selection was competitive: I was selected as 1 of 500 participants nationwide from 12,176 applicants (a 4.1% acceptance rate) through a multi-stage evaluation assessing technical programming proficiency, digital problem-solving, and communication skills.
The BigBox Data Ingestion Problem
For our capstone challenge, our cross-functional team of five (two developers, two UI/UX designers, and one product manager) was assigned to solve an operational bottleneck for BigBox, Telkom Indonesia's enterprise big data platform.
BigBox regularly onboarded external enterprise client data arriving in messy, nested spreadsheet formats (XLS and XLSX). Because ingestion pipelines require clean, structured JSON schemas, technical staff had to write ad-hoc Python conversion scripts for every single client dataset. The manual process was slow, error-prone, and tied up engineering hours that should have been spent building platform features.
Our initial prototype attempted to parse client workbooks entirely in the browser using standard JavaScript file readers. That immediately hit a technical wall: when handling large enterprise files (often 30MB to 50MB with multiple nested worksheets and inconsistent column headers), parsing the raw binary into memory locked the browser's main thread, causing painful UI freezes and Out-Of-Memory crashes.
Building BigOne Ultimate Converter
To eliminate browser crashes while keeping client data secure without sending sensitive raw spreadsheets to external servers, we engineered BigOne Ultimate Converter in React.js.
As the frontend developer on the Hacker track, I solved the performance bottleneck by architecting a client-side chunking and streaming parser:
- Asynchronous chunked streaming: Instead of loading entire multi-megabyte workbooks into memory simultaneously, the parser slices worksheets into discrete record batches, keeping the browser UI smooth and responsive at 60 FPS.
- Automated schema inference: The engine inspects row samples to infer data types (dates, booleans, floating-point numbers, ISO timestamps) and flags malformed or empty cells for instant user review.
- Side-by-side visual mapping canvas: Non-technical data operators can drag and drop schema keys, rename object properties, nest related arrays, and preview the live JSON tree in real time before export.
- Validated output export: One-click export produces minified or formatted JSON schemas calibrated to plug straight into BigBox ingestion endpoints.
National Pitching & 1st Place Award
The project advanced through competitive internal reviews before reaching the national finals, where our team pitched live before senior TelkomGroup executives.
The judging panel was chaired by Muhamad Fajrin Rasyid, Chief Digital and Information Officer (CDIO) of TelkomGroup and co-founder of Bukalapak. We demonstrated the live converter, walked through our async streaming architecture, and proved how the tool reduced enterprise client onboarding time from several hours of manual Python scripting down to under three minutes.
BigOne Ultimate Converter was awarded 1st Place (TOP #1 Best Project Team) nationwide out of more than 100 project teams.
Tech Stack & Skills
- React.js
- Frontend Development
- UI/UX Design
- Team Leadership
- Product Pitching