Practical AI systems, end to end
I build production AI workflows : OCR document pipelines, RAG-based knowledge retrieval, LLM automation, and monitoring dashboards. Each project starts from a real bottleneck in the field, not from wanting to use a particular model.
My approach: find where manual workflow slows things down, then decide if AI is the right fix. The result is systems that are simple to run, easy to monitor, and don't create more work than they solve.
Indonesia (GMT+7) · AI Engineer
Photo coming soon
Projects are framed around messy inputs, review states, user decisions, and handoff points.
Coverage from ingestion and model calls to APIs, databases, dashboards, deployment, and monitoring.
Case studies include trade-offs, failure modes, validation paths, and what should improve next.
Every system ships with logs, evaluation samples, and human override paths built in, not bolted on.
Data flow, failure modes, and validation rules are designed before any model is chosen or UI is built.
Each project lands as a minimal production-ready slice — monitored, documented, and ready to improve.
Featured work
Each project explains the problem, architecture, AI component, data flow, trade-offs, and what still needs real-world validation.
OCR / Document Processing
A FastAPI + Next.js service that extracts structured data from invoices, receipts, and tax documents using PaddleOCR.
Problem Solved
Finance teams manually key in data from invoices and receipts. This pipeline turns scans into structured JSON with a review UI.
LLM Workflow / HR Automation
A full-stack AI screening tool that parses CVs, maps them to job requirements, and ranks candidates : with transparent evidence for every score.
Problem Solved
Recruiters spend too much time reading CVs without a consistent rubric. This tool gives them structured insight per candidate, mapped to the job description.
Computer Vision / Field Operations
A face recognition attendance system for multi-outlet field staff, with real-time dashboard and location verification.
Problem Solved
Field supervisors need to verify staff attendance across multiple outlets without being physically present at every location.
Skill snapshot
The stack is organized around what a real AI system needs: model capability, data flow, application logic, infrastructure, and operator-facing UX.
Proof of engineering thinking
Can model a business workflow before choosing tools.
Can design RAG, OCR, and computer vision pipelines with review loops.
Can connect AI services to backend APIs, databases, and dashboards.
Can communicate technical trade-offs clearly to recruiters, CTOs, and clients.
Available for focused AI builds
I can help design the pipeline, backend, dashboard, and production checks needed to turn an AI idea into a maintainable system.