AI/ML Engineer · Jakarta
Production NLP and machine learning, grounded in years shipping real systems, from social listening to market intelligence.
I'm an engineer who likes the unglamorous part of AI: making it work reliably inside real products. Most of my time goes to language and data: reading the mess, finding what matters, and shipping it as something people can actually use.
Now I work on my own, putting that same care into the products I build.
Sentiment, classification, and entity resolution pipelines that process large volumes of messy, real-world text, reliably, not just in a notebook.
Structured output, enrichment, and semantic search wired into products, solving concrete problems instead of chasing hype.
Embeddings, vector search, ranking, and record linkage: the quiet machinery behind good recommendations and clean data.
Cloud-native pipelines on BigQuery and Cloud Run: serverless, event-driven, and cheap to run at scale.
Built as the technical backbone of a data science team at a media intelligence company, turning raw text into systems analysts use every day.
Producing one report meant six-plus hours of pulling data, running the numbers, and writing it all up. I built an LLM pipeline that does the whole thing, from raw data to a finished report, and reruns on a schedule without adding any human load.
6+ hrs of manual work → under 1 minute, recurring
Not every spike is a crisis. Together with the team, I worked on a monitoring system that watches selected posts and escalates them through four stages, Normal, Watch, Hot, Crisis, by reading engagement velocity, sentiment, and volume together, so an early crisis surfaces before it grows while the noise stays quiet.
An AI assistant is only as good as the data behind it. I owned the backend, designing the services and data layer that fed clean, structured, query-ready data to the analyst-facing AI assistants, so they answered from solid ground instead of raw mess.