I'm Ragil Hadi Prasetyo, an AI engineer at GLAIR | GDP Labs. I work on document processing and retrieval for RAG applications, and OCR for Indonesian documents like KTP. On the side, I build small developer tools to smooth out everyday engineering problems.
I'm an AI engineer who likes the practical side of the work — getting messy documents into a shape that models and applications can actually use.
Based in Jakarta, I work at GLAIR | GDP Labs on document processing and retrieval for RAG applications, and earlier on OCR for Indonesian documents like KTP. Most of what I do is on the document and data side: parsing, structuring, and making things searchable and reliable.
Outside of that, I enjoy building small developer tools to fix recurring annoyances in everyday engineering. Before this I mentored aspiring data scientists, which is still how I like to think — clearly, from the fundamentals up.
Work on document processing and retrieval that powers RAG applications — parsing, structuring, and indexing documents so they can be searched and answered against reliably. Focused on the document and data side of things.
Built OCR for Indonesian documents — handling low-quality scans, varied layouts, and reliable field extraction — along with a document-quality detector that catches bad captures before they reach OCR. Also worked on supporting vision and NLP services for client projects.
Mentored students through the full data-science curriculum — Python, statistics, machine learning, and deployment. Designed assignments and reviewed projects, reinforcing the fundamentals I rely on today.
Developer infrastructure, terminal tools, and interactive web products — built for real workflows and explained beyond the headline.
A terminal-first, provider-neutral coding-agent platform with guarded tools and resumable sessions.
A cross-platform alias manager that turns one portable store into native commands across your shells.
Generates real commands for bash, zsh, fish, and PowerShell while the CLI runner works in scripts, CI, and unsupported shells.
Supports positional and named placeholders, defaults, all-argument forms, safe quoting, and extra argument forwarding.
A safety-first machine audit tool for finding forgotten files, caches, containers, and development artifacts.
Scanners cover Docker, Python, Node, Rust, Go, JVM tools, browsers, editors, downloads, screenshots, temporary files, and Git branches.
Audit and plan never change the filesystem, while clean remains a preview until an explicit --apply.
A self-hosted URL shortener with a Rust CLI, HTTP API, web interface, expiration, and visit analytics.
Create a short link from one terminal command, choose a custom server, pass a custom code, or set a time-to-live.
Supports generated or custom codes and expiration from five minutes to thirty days.
A container-native load tester for repeatable YAML scenarios, live metrics, quality gates, and portable reports.
Tokio-based async workers maximize throughput while sync mode supports controlled rate-limit testing.
JSONPath extraction and variable templates connect login, profile, upload, and other dependent request flows.
A fast, lightweight HTTP mock server that turns simple JSON files into realistic API behavior.
One readable file can define an endpoint’s method, path, status, body, headers, and request-consumption behavior.
Matches path parameters, query parameters, headers, and JSON, text, or form bodies—not only an exact URL.
An interactive mathematics web app that makes geometry concepts visual, direct, and easier to explore.
Geometry is presented as an interactive visual subject instead of a sequence of static formulas.
The browser experience encourages learners to explore relationships and build intuition through interaction.
Training, evaluation, and inference for vision & deep-learning models.
Grounded, evaluated language applications and RAG pipelines.
Fast, typed services that put models in front of real traffic.
Reproducible training and safe, observable deployments.
Occasional writing on machine learning and computer vision — full posts live on Medium.
Discover how content-based filtering works in recommendation systems. This article walks through the core concepts and implementation steps to build a movie recommendation engine using machine learning.
Learn how to build a simple face recognition system by combining facial landmarks with the K-Nearest Neighbors algorithm. A beginner-friendly guide to applying computer vision and machine learning for facial recognition.
Open to AI engineering roles, consulting, and collaborations. I usually reply within a day.
[email protected]