Applied AI product engineer, Indonesia
Production systems that connect language models to real company knowledge, business rules, and the people accountable for the outcome. Operations, knowledge work, commerce, and vertical software.
About
I'm an applied AI product engineer based in Indonesia. I design and build the layer between language models and real operations: retrieval, business rules, review workflows, and the backend that keeps it all running. My background is in full-stack product engineering, which is why I care as much about queues, audit trails, and cost as I do about the model.
Every system I build shares one shape. Company knowledge is retrieved and assembled into context, Claude reasons over it, and anything consequential pauses for a person before it becomes an action. Each run leaves a trail you can audit.
Capability-led, not client-led: these describe the kinds of systems I build. Open a row for the details.
A retrieval-augmented assistant for operational teams. It answers only from approved sources, routes requests to the right queue, automates the repetitive steps, and escalates to a person when the situation calls for judgment rather than a confident guess.
Systems that ingest policies, structured data, and operational context, then answer with citations instead of generic model output. Retrieval quality, chunking strategy, and freshness are treated as product features, not plumbing.
Structured AI pipelines where consistency matters more than novelty: research, extraction, transformation, and drafting steps with schemas, evaluation checkpoints, and clear rules for what gets published automatically and what waits for review.
Reusable foundations for AI-enabled products: queues, model orchestration, retrieval, audit trails, human review checkpoints, API integration, observability, and cost-aware processing. The unglamorous layer that decides whether a prototype becomes a product.
I work across the whole implementation path, so the hand-offs that usually break AI projects are mine to own.
LLM product design, prompt systems, structured outputs, retrieval-augmented generation, and model integration.
Document ingestion, embeddings, retrieval, source-aware context, and operational knowledge layers.
Multi-step automation with human checkpoints, business rules, queues, APIs, and production monitoring.
Full-stack systems with Laravel, Livewire, Filament, Next.js, relational databases, Redis, and cloud infrastructure.
My next phase of work focuses on Claude-powered systems for long-context understanding, grounded retrieval, structured generation, document-heavy workflows, and reliable multi-step assistance. Pick a stage to see how the pieces fit.
Long-context reasoning over the assembled material: answering with citations, extracting structured fields, drafting, or deciding which step comes next. Output is shaped with schemas so downstream code can trust it.
AI should remove real operational friction. A demo that generates text is not the same as a system that closes tickets.
The model is only as good as what it is shown. Approved data, source material, business rules, and the current task come first.
Consequential steps get review boundaries, observable outputs, and deliberate approval. Automation earns trust gradually.
Reliability, cost, latency, queues, security, and evaluation decide whether it survives. The prototype is the easy part.
If you are building an AI-native product where proprietary knowledge, structured workflows, and human expertise matter, I'd like to hear about it.
Start a conversation