Irsyad Zulfikar

Architected a production smartphone-grading pipeline of six detection, classification, and segmentation models. Led the migration from Docker Swarm to GKE, tuned per-service autoscaling, built Vertex AI training pipelines with MLflow, and added inference monitoring with VictoriaMetrics and Grafana. Benchmarked 10,000+ samples across multiple YOLO generations and shipped real-time OCR for paperless trading-machine transactions.

Built an LLM-powered CV parsing and scoring pipeline with LlamaIndex and Groq, reducing manual screening time by 50%. Developed RAG assistants with Flask and MongoDB, a live coding assessment platform, and recruitment analytics pipelines for HR teams.

Developed vehicle detection and counting models with 86% validation accuracy and a waste-classification CNN with 82% precision on hazardous objects. Defined annotation and preprocessing standards to improve generalization across varied data conditions.