Nahli Robbani

Electrical & AI Systems Engineer with 1+ years of hands-on experience in Edge AI, Autonomous Agentic Workflows, and Production Data Engineering. Adept at bridging hardware and software gaps, from deploying CUDA-accelerated YOLO models on resource-constrained NVIDIA Jetson hardware to architecting hybrid Text-to-SQL/RAG pipelines on self-hosted PostgreSQL. Track record includes boosting SOP compliance by 35% at Toyota (TMMIN), reducing data retrieval latency by 60–70% for TNI AD, and launching scalable health-tech platforms with sub-2s inference speeds.

Automation and Data Engineer at PT Global Data Inspirasi (2025-12 – Present)

• Architected and deployed an autonomous Text-to-SQL & Hybrid RAG Agent in n8n, orchestrating Groq LLM, Supabase Vector Stores, and dynamic PostgreSQL tools to query structured military databases (logistik_transaksi, personel) alongside unstructured SOPs.
• Engineered dynamic schema-inspection tools and strict SQL parsing prompt-rules (handling CamelCase identifier quotes, LIKE wildcard matching, and multi-table joins) to execute zero-error database transactions autonomously.
• Designed and optimized 3NF PostgreSQL schemas for defense logistics, enforcing relational integrity via dynamic foreign keys (ON DELETE CASCADE/SET NULL) across military personnel (NRP) and asset allocation tracking.
• Eliminated manual SQL query writing for end-users by enabling natural language database querying with 100% schema compliance and zero syntax execution errors.
• Boosted military data retrieval accuracy by over 40% by combining structured SQL execution with semantic vector search (sop_tni_jumbo).
• Successfully structured and embedded 500+ military files into Supabase vector stores, providing instant real-time knowledge retrieval for field operators.

AI System Architect at PT Vuriko Developer Studio (2025-08 – Present)

• Architected and deployed multi-agent AI triage workflows in n8n using LangChain Agents & Gemini 2.5 Flash Lite to classify unstructured user data into specialized medical-risk pathways (Special Case vs General).
• Engineered hybrid decision pipelines inside n8n, combining deterministic JavaScript execution (BMI indexing, joint-load risk scoring) with dynamic prompt injection for automated fitness diagnostics.
• Built and maintained self-hosted Docker server infrastructure to handle asynchronous webhooks and automated container update pipelines across all deployed AI services.
• Optimized end-to-end AI agent reasoning latency to 0.5–2.0 seconds by deploying lightweight LLMs and stream-lined n8n data transformation nodes.
• Successfully co-founded and launched the Mindfit digital platform, fully automating initial health-tech analytics and patient triaging without manual intervention.
• Maintained 99.9% system uptime across all self-hosted Docker server infrastructures and n8n automation pipelines.

Computer Vision Intern at PT Toyota Motor Manufacturing Indonesia (2025-04 – 2025-07)

• Engineered and deployed an edge-AI compliance monitoring system on NVIDIA Jetson Nano using YOLO pose estimation, tracking wrist-to-shoulder kinematic vectors and ROI polygon intersections to enforce operator SPC protocols.
• Developed an industrial PPE compliance pipeline using a custom-trained YOLO model with CUDA acceleration, implementing dynamic letterbox aspect-ratio preprocessing and NMS threshold tuning for automated real-time helmet detection.
• Architected custom multi-frame trajectory logic and direction vectors (dx/dy) to track movement patterns, preventing false positives during operator entry and exit transitions.
• Integrated closed-loop hardware actuation via Jetson GPIO (traffic light output) and asynchronous Pygame audio cues to trigger physical alerts upon protocol violations.
• Boosted manufacturing safety crosswalk compliance by 35% through real-time kinematic pose monitoring in high-risk zones.
• Maintained a stable 10 FPS inference speed and 80%+ accuracy for pose-estimation models deployed on resource-constrained NVIDIA Jetson Nano edge hardware.
• Achieved 30+ FPS real-time throughput on custom CUDA-accelerated YOLO helmet detection pipelines for automated PPE safety compliance.

Research Assistant (Computer Vision) at Universitas Negeri Semarang (2023-12 – 2024-12)

• Led the development of a multi-model computer vision pipeline using Keras, TensorFlow, and OpenCV to perform real-time facial attribute analysis (age, gender, and 7-class emotion recognition).
• Architected multi-input preprocessing workflows, dynamically cropping and scaling ROI layers (48x48 grayscale & 200x200 RGB) to feed three parallel CNN inference networks simultaneously.
• Benchmarked edge AI hardware performance, evaluating Keras/TensorFlow model execution speeds to analyze throughput gains between NVIDIA Jetson Nano and Jetson Xavier AGX.
• Engineered a multi-task deep learning pipeline using OpenCV and Keras, executing 3 parallel CNN models for real-time age estimation, gender classification, and 7-class emotion recognition.
• Optimized multi-input image preprocessing workflows by dynamically cropping and scaling ROI layers (48x48 grayscale & 200x200 RGB) to sustain continuous live webcam inference loops.
• Benchmarked edge AI hardware throughput, proving NVIDIA Jetson Xavier AGX delivers up to 300% faster inference performance (sustaining 20–25 FPS) compared to Jetson Nano under multi-model workloads.

Bachelor of Engineering (B.E.) in Electrical Engineering – Universitas Negeri Semarang (2022 – 2026)