Hi — I'm Sevian.

§ At Luce SG, these days

For the past year and change, the work has cut ~70% off backend latency through Rails service and query refactors, halved CI from 45m to 15m, and lifted monitoring coverage by ~50%. Luce is recognized by the Financial Times as one of Asia-Pacific's fastest-growing companies.

The flagship project is the autopay failure retry & escalation system — payment-attempt tracking, date-based retry logic, Stripe error translation, auto-disable, and a customer notification pipeline — designed end-to-end. It recovers revenue across the billing lifecycle that would otherwise quietly leak.

The cross-repo bet was an AI-powered sales automation spanning backend, the communication platform, and the AI-agent service. It identifies cold leads, sends contextual follow-ups, and progresses deal stages on its own. Alongside it, I designed the orchestration for the customer-facing AI assistant and its triage layer — the routing logic that decides AI-vs-human handling, plus the state-consistency wiring across Twilio and HubSpot that keeps the handoff coherent in either direction.

Beyond those: credit-based bookings rolled out across service verticals, a local-first HubSpot contacts cache, and structural work in the communication platform — rebalancing the sync/async boundary, consolidating concurrency control onto Postgres, and capping memory on the conversation-history path.

§ Before Luce — Ukirama, '23–'24

A cloud SaaS ERP covering inventory, finance, and core business workflows — clients report roughly 75% faster reporting through the automation it unlocks. I architected and built the core approval flow engine from the ground up — complex, customizable workflows sitting right at the heart of the product.

The work I'm most proud of there: sped ERP reports up by ~50% by migrating a key internal library from Ruby to Go, and built a monitoring & health-check pipeline on AppSignal so we could flag anomalies before users did.

§ Earlier — Orami & Larona, '19–'23

At Orami (part of SIRCLO's e-commerce ecosystem), I cut incident resolution time by ~90% by writing runbooks and documenting recurring failure patterns — boring, unglamorous work that made the whole team faster. I also stabilized PostgreSQL and MySQL under high-traffic e-commerce workloads.

Before that, at Larona Prima Solusi, I built an offline-first SPA for low-connectivity ERP usage and wrote C++ integrations connecting client ERPs with marketplace APIs. First real job, first real taste of shipping to customers whose internet didn't always cooperate.

§ On the side

I maintain a couple of small things that solve annoyances I actually had.

event_timeline ↗ is a reusable audit library for Rails. It tracks model state changes as structured event trails — the kind of thing you wish you had when a customer asks, a month later, exactly what happened to their record.

redacted ↗ is a secret-redacting proxy for AI coding assistants. It detects sensitive strings and swaps them for [REDACTED] before anything reaches an external API — so I stop accidentally pasting credentials into prompts. And maybe you stop too.

§ The toolbox

Ruby and Go at home; Rails and React / React Native daily; Python, TypeScript, C++, plus AL when Business Central asks. Postgres, MySQL, and Redis for data. Docker, Sidekiq, CircleCI, RSpec; observability through Sentry, NewRelic, AppSignal.

B.Sc. Information Systems from Universitas Bunda Mulia, 2018 — 2022.