Marcellinus Santoso

SUMMARY I'm an AI engineer, 12 years in, and back to writing code daily to keep pace with LLMs. Seven of those years were forward deployed engineer work in a software house I co-founded in Jakarta, taking custom systems for enterprise and government clients from a vague first conversation through to production support. Today I lead technology at The Body Shop Indonesia, where the six generative AI systems I built run on multi-agent orchestration, MCP servers, RAG and evals, in production across seven of ten business functions. I studied in Singapore and began my career there, writing code on a SGD $7M government mobile project.

EXPERIENCE The Body Shop Indonesia, Jakarta Head of Technology · Aug 2025 - Present Embedded with finance, marketing, analytics and brand teams: scoped agents to their workflows, owned delivery to adoption. Built the platform: multi-agent orchestration on a self-built harness (Hermes), MCP servers, RAG over PostgreSQL/pgvector, an Obsidian knowledge graph, LangChain, n8n, OpenRouter. Rebuilt how engineering ships: spec, implementation, eval, human review every change. Evals drive prompt iteration; guardrails block unreviewed writes to production. Rolled the analytics agent out to 5 departments queued behind one analyst. Analysis: 3 days to about 10 minutes; Finance P&L, a full working day to roughly 5 minutes. Moved services off TypeScript to Go, cutting the infra bill about 80%. Cloud infrastructure agents monitor and maintain about 98% of the hybrid on-premise/AWS estate: 20 services, 300 Kubernetes pods, no dedicated devops hire. Own the system prompts, tool schemas and retrieval layer. Tune the agent loop and inference cost: context budget, tool-call limits, stopping conditions, model routing across providers. IT Manager, Digital Transformation · Mar 2022 - Jul 2025 Led a 3-year in-house rebuild of core retail systems off vendor platforms: POS, e-commerce, OMS, marketplace, loyalty, monitoring. Scoped each with its business owner. Migrated legacy Java and PHP to TypeScript (Next.js, NestJS), MongoDB, Flutter, Kubernetes microservices, CI/CD: page load 20 seconds to under 1 second. Built in-house instead of buying per-seat licences. Platform cost down about 95% with a leaner team; change cycles from 2-3 months to about one week. Built the first automations and agent prototypes here. They grew into the platform above.

Holywings Indonesia, Jakarta Head of IT · Aug 2021 - Feb 2022 Rebuilt three legacy apps in Laravel and Flutter with CI/CD. Led a team of 15 developers.

CODINA.IO, Jakarta Quantitative Analyst & Co-Founder · Feb 2019 - Aug 2021 Co-founded it, wrote a fully automated crypto trading bot in Python: ran live for two years unattended, managing USD 150,000.

INDO MEGA BYTE, Jakarta · own venture, deliberately run alongside the employed roles above Co-Founder & Full-Stack Developer · Sep 2014 - Aug 2021 Ran the software house: scoped and built mobile and web apps for around 15 enterprise and government clients, including The Body Shop and BMKG. Forward deployed engineer work under a different title: technical discovery, solutions engineering against a vague brief, then the code. Stayed on every engagement past customer deployment: incident response, change requests; whoever promised the system got the 2am call. Explained trade-offs to non-technical stakeholders in their own terms; The Body Shop started as a client here, later a full-time role.

iAPPS Pte. Ltd., Singapore Software Developer · 2013 Wrote production code on a SGD $7M Singapore government mobile project.

AI SYSTEMS IN PRODUCTION Cloud infrastructure agents monitor, deploy and maintain the hybrid on-premise and AWS estate. Data analytics and data warehouse agents turn business questions into SQL. AdOps agents run day-to-day advertising operations for the marketing team. Market research agents synthesise category and competitor research. Finance agents automate recurring finance workflows and reporting. Brand systems agent generates on-brand assets via hybrid search and document chunking.

SKILLS Languages: JavaScript/TypeScript (Next.js, NestJS), Go, Python, SQL AI engineering: Large language model (LLM) application development, generative AI systems, multi-agent orchestration, agent harness design, agent skills, RAG, hybrid & vector search, knowledge graphs, evals, guardrails, system-prompt design, loop engineering, MCP servers AI coding agents: Claude Code, Codex, Augment Code, Grok Build, GLM Coding Data & infra: PostgreSQL (incl. pgvector), MongoDB, Redis, Kafka, Elasticsearch, Kubernetes, AWS, microservices, CI/CD Tools: LangChain, LangGraph, LangSmith, n8n, OpenRouter, Hermes Agent, Flutter, Laravel

EDUCATION James Cook University Singapore. Bachelor of Information Technology, 2011

REFERENCES Available on request.