AI & Software Engineer — GMT+7

Low-latency, resilient, and cost-efficient systems.

I ship products where infra OpEx and reliability are features. Swift low-level to edge serverless — strict validation, deterministic errors, zero main-thread blocking.

Robustness strictFail-closed schemas & POSIX 0600
Resilience fastCarbon C-APIs & async event-loops
Prod Ready OpExDynamoDB GSI & Edge Workers

Architecture prioritized for accuracy, non-blocking execution, and scalable OpEx — verified through Instruments profiling and strict schema validation.

Selected work — architecture & outcomes

3 public live, 1 private. Each with real challenge → decision → outcome.

Public SaaS — fresworkly.com
Next.js 15 RSCReact 19DynamoDB GSIVercel AI SDK
● Live
Challenge

Real-time trending feed under high concurrency without database lock contention.

Decision

Exponential half-life decay T½=30m, TTL 90m computed in memory, partitioned dual-tier auth (Better Auth + isolated JWT admin) + Edge WAF allowlist.

Outcome → user impact

Lock-free trending updates, instant feed evaluation, and predictable DynamoDB multi-table GSI scaling.

Next.js 15 · DynamoDB multi-table GSIOpen live ↗
macOS Launcher — rouser.my.id
Swift 5.9Carbon C-APIsmdfindSparkle
● Live
Challenge

Intercepting global hotkeys instantly while preventing Spotlight AX paste operations from stealing window focus.

Decision

Carbon C-APIs for hotkey interception, Ifrit event-driven fuzzy aggregator, lazy AXUIElement with Toast queue fallback, [weak self] strict safety.

Outcome

Instant hotkey response path on macOS, 0 main-thread blocks traced via Instruments, clean LSUIElement daemon background footprint.

SwiftUI+AppKit · Offline JSONOpen live ↗
Edge App — corerain.biz.id
Swift ActorsPOSIX cpulimitCF WorkersD1 SQLite
● Live
Challenge

Fine-grained CPU throttling without kernel instability, paired with zero-server cost offline licensing.

Decision

POSIX SIGSTOP/SIGCONT with proc_pidinfo identity, 3-state license machine, file security 0600, Workers + D1 edge backend.

Outcome

Stable CPU limits across target ranges, zero-infra licensing cost, instant global edge distribution, offline license verification.

Edge · $0 infra · POSiXOpen live ↗
Enterprise Internal — Confidential
Nexus Curation & Fast Crawl Platform
Python 3.12FastAPITaskiq + RedisPostgreSQLSSE
Private
Challenge

High-volume multi-platform post ingestion across social handles, handling dynamic rate-limits, anti-bot blocking, and cross-platform handle duplication.

Decision

Pydantic v2 strict validation, Taskiq + Redis distributed workers, SSE heartbeat 2s to avoid CF 524 on LLM streaming, aioboto3 async S3 + local auto-cleanup, smart handle deduplication.

Outcome

Fast incremental daily crawl cycles, significant bandwidth savings via deduplication, zero disk bloat.

1 screenshot only (WebP 34KB) to keep page <80KB. Full walkthrough under NDA.
Nexus Curation Queue showing pending posts, smart deduplication notice, filters for vendor and platform
Curation Queue — Active Queue View↗ zoom • WebP
Python 3.12 · FastAPI · Taskiq + Redis · 14 slot WIB scheduleNDA walkthrough available

Principles — 2026

Separating prod from demo.

01 — Robustness

Strict by default

Partitioned auth tiers, WAF allowlist, Pydantic v2, SQLAlchemy 2.0, POSIX 0600. Fail closed.

02 — Resilience

Fast & non-blocking

Low-level C-APIs, SIGSTOP/CONT process control, SSE heartbeats, async S3 storage.

03 — Production

OpEx is a feature

Serverless edge architectures, atomic GSI writes, LSUIElement daemon, offline-first design.

Stack — minimal, high leverage

Chosen for latency, cost, DX.

Systems & Languages

  • Swift 5.9 Actorsnative
  • Goprod
  • Python 3.12 FastAPIprod
  • TypeScript 5 / Next.js 15prod
  • POSIX C / Carbonlow-level

Data & Infra

  • DynamoDB GSIserverless
  • PostgreSQL 16txn
  • Redis + Taskiqqueue
  • Cloudflare Workers / D1edge
  • S3 aioboto3 asyncstorage

Patterns

  • Event-Driven Ifritresilient
  • SIGSTOP/CONT throttlingcontrol
  • Async Queues + SSEthroughput
  • WAF + Partitioned Authrobust
  • OpEx modelingready