Five years building iOS for
Indonesia's largest super app.

Post-COVID, Tokopedia identified medicine sales as a major untapped market. I led the end-to-end iOS initiative — from business research and vendor licensing strategy through to the full purchase flow.

The core challenge: medicines require pharmacy vendor licensing, and many require a valid prescription. We built a catalog classifying every medicine as OTC or prescription-only, then wired up the entire compliance layer in-app.

What we built

• ▸ Medicine catalog with OTC vs. prescription classification
• ▸ In-cart prescription flag — prompts upload when required
• ▸ Backend OCR validation of prescription images
• ▸ Consultation module via HelloDoc (on-call doctor, free tier)
• ▸ Prescription-to-checkout bridge — validated script unlocks purchase
• ▸ Vendor licensing strategy via onboarded pharmacy partners

Tokopedia had always required upfront payment. BNPL introduced deferred payment — a first for the platform — and I owned the entire iOS implementation across every entry point.

The feature was driven by product value, not product type: given the item price, we surface available lenders, tenures, and charges side by side so the customer can compare and choose before adding to cart.

Implementation

• ▸ PDP widget — lender comparison, tenure options, charge breakdown
• ▸ Cart widget — selected BNPL option persisted through checkout
• ▸ Cart API extension — additional fields passed with BNPL selection
• ▸ Multi-provider support with sort by tenure or interest rate

I proposed the Apple Watch idea, designed the initial UX wireframes, and built the extension end-to-end. The watch app surfaces a user's wishlist items — product name, price, and a deep link back to the main app.

The feature was featured on the Indonesian App Store — a milestone for the team.

Technical challenge

The main app used a static network library to reduce compile time. Adding the Watch extension meant the static library was bundled twice — once in each target — causing significant binary size bloat.

The fix: convert the network module from static to dynamic library. Both targets then shared a single copy at runtime, eliminating the duplication and bringing bundle size back under control.

Every payment flow in Tokopedia — product purchase, seller wallet withdrawal, top-up, and deferred ATM payments — lands on the same terminal page. I rebuilt this as a single, signal-driven surface that adapts its layout entirely based on payment method.

Deferred transactions (common in Indonesian banking) show a live countdown and payment instructions. The rating engine and product recommendations were added as optional layout slots.

The layout-driven pattern

A single API returns layout definitions for a page. A second pass fetches component data only for visible layouts — keeping payloads small and enabling streaming. If a component fails, it is hidden silently. If a layout fails, a targeted retry fetches only that layout ID.

This pattern scaled beyond the Thank You page — it became the standard for brand pages, product recommendations, and every new server-driven surface we built after.

Tokopedia's AppDelegate was a monolith that nobody wanted to touch. Universal links, deep linking, analytics, A/B experiments, push notifications — years of accumulated responsibility in a single file. One mistake would break the app at launch.

I led the migration in phases: extract each responsibility into its own manager class with mirrored lifecycle methods, then make AppDelegate a pure orchestrator. Final step: rewrite in Swift.

Safe rollout strategy

• ▸ Phase 1: Extract analytics, deep link router, A/B testing, push notifications into separate manager classes
• ▸ Unit tests asserting each lifecycle method is called exactly once
• ▸ New Swift AppDelegate shipped to beta group across 4 release cycles
• ▸ Production flipped only after beta validation — zero incidents