CV

Curriculum Vitae detailing academic trajectory, systems engineering projects, and research at Institut Teknologi Sepuluh Nopember (ITS).

Contact Information

Name Abdurrahman Izzuddin Al Faruq
Professional Title Information Systems Undergraduate & AI Systems Researcher
Email fixgmc@gmail.com
Location Departemen Sistem Informasi, FTEIC - ITS, Surabaya, Jawa Timur 60111

Professional Summary

Undergraduate student in Information Systems at Institut Teknologi Sepuluh Nopember (ITS) Surabaya and active researcher at Laboratorium Analisis Data dan Diagnostik Informasi (ADDI). Specializing in Machine Learning Systems, CUDA kernel engineering for NVIDIA Blackwell architectures, Large Language Model Alignment (DPO vs. PPO on statutory legal corpora), and Enterprise Information Systems Architecture.

Experience

  • 2026 - Present

    Surabaya, Indonesia

    AI Engineer (Part-Time)
    Unit Pengelolaan dan Pengendalian Program (UP3), ITS
    Engineering enterprise agentic RAG systems, custom Model Context Protocol (MCP) data services, and high-throughput sovereign LLM inference runtimes.
    • Architected and deployed an internal enterprise Agentic RAG system with a conversational interface for administrative intelligence and program analytics.
    • Developed a custom Model Context Protocol (MCP) server enabling secure, low-latency agentic query synthesis over large-scale institutional MS SQL Server databases.
    • Self-hosted sovereign foundation models (Qwen 27B–35B) using optimized bare-metal inference runtimes, achieving sustained decode throughput of up to 1,300 tokens/sec.
  • 2025 - Present

    Surabaya, Indonesia

    Undergraduate Thesis & AI Alignment Researcher
    Laboratorium Analisis Data dan Diagnostik Informasi (ADDI), ITS
    Conducting empirical alignment and post-training research on statutory legal reasoning and digital sovereignty.
    • Designed and implemented empirical alignment experiments comparing Direct Preference Optimization (DPO) and Proximal Policy Optimization (PPO) for Indonesian Criminal Code reasoning.
    • Curated and validated specialized legal preference datasets and conducted rigorous SFT and alignment benchmarks on open-weight LLMs (Qwen-series).
    • Employed vLLM candidate generation and reward model scoring to evaluate legal fidelity and reduce statutory hallucination.
  • 2024 - Present

    Surabaya, Indonesia

    AI Systems & Compiler Engineer
    Systems & High-Performance Computing (Open Source)
    Engineering bare-metal C++/CUDA inference runtimes and distributed reinforcement learning pipelines.
    • Developed NInfer: A single-GPU bare-metal C++/CUDA inference engine targeting NVIDIA Blackwell architecture (RTX 5090 & RTX 5060 Ti) with exact-batch CUDA Graphs and MTP speculative decoding.
    • Engineered AuK Optimization: Bare-metal streaming engine for Tencent Hunyuan AuK-Flash 1.5B audio foundation model, delivering 30.4x real-time throughput and sub-100ms TTFA via custom CuTe SM120 BF16 GEMM.
    • Created Yoshida: Production-grade General LLM Reinforcement Learning & Post-Training framework supporting PPO, DPO, and preference-tuning workflows.
    • Authored Axon: Minimal deep learning framework written in C++20 from scratch with eager autograd, AVX2 SIMD CPU backend, and custom quantization.

Education

  • 2024 - Present

    Surabaya, Indonesia

    Bachelor of Science / Sarjana Komputer (S-1)
    Institut Teknologi Sepuluh Nopember (ITS)
    Sistem Informasi (Department of Information Systems)
    • Undergraduate Thesis Researcher at Laboratorium Analisis Data dan Diagnostik Informasi (ADDI) focusing on Legal LLM Alignment (DPO vs PPO on statutory legal reasoning).
    • Specializing in Machine Learning Systems, bare-metal GPU inference, and high-performance CUDA computing.
    • Active in open-source AI infrastructure and deep learning systems research.

Skills

AI & Deep Learning Systems (Advanced): PyTorch, CUDA C++, CuTe / CUTLASS, vLLM, TensorRT-LLM, Agentic RAG, Hugging Face (TRL, Transformers, Accelerate), RLHF, DPO, PPO, Generative Flow Matching
Software & Systems Programming (Advanced): C++20, Python 3, CUDA, SQL (MS SQL Server, PostgreSQL, MySQL), Model Context Protocol (MCP), Bash / Shell scripting, Linux Systems Programming, AVX2 SIMD
Systems Engineering & Infrastructure (Proficient): Git, GitHub Actions CI/CD, Docker, Linux Systems Administration, NGINX, Distributed Training Cluster, Slurm, vLLM Engine

Projects

  • NInfer (RTX 5060 Ti)

    High-performance C++/CUDA single-GPU inference engine for NVIDIA Blackwell architecture.

    • Custom CUDA kernels for SM120 tensor cores, NVFP4 / INT8 quantization, and multi-token prediction speculative decoding.
    • Warmed-up steady-state execution with exact-batch CUDA Graph capture, achieving deterministic low latency.
  • AuK Optimization (RTX 5060 Ti)

    Bare-metal inference & low-latency audio streaming engine on Blackwell SM120.

    • Achieved 24.3x to 30.4x real-time factor and sub-100ms Time-To-First-Audio without runtime UI overhead.
    • Custom CuTe BF16 GEMM & Fused SwiGLU operations eliminating FP32 intermediate register staging.
  • Yoshida Framework

    General LLM Reinforcement Learning & Post-Training Framework.

    • Modular post-training engine supporting full-epoch SFT, DPO, and PPO candidate scoring for research reproducibility.
  • Axon Deep Learning Framework

    Minimal deep learning framework written in C++20 from scratch.

    • Dynamic graph autograd engine with vectorized AVX2 kernels and GGML-style quantization.

Languages

Bahasa Indonesia : Native speaker
English : Professional working proficiency (Technical reading, writing, and documentation)

Interests

Machine Learning Systems: High-Performance Computing, GPU Architecture, CUDA Optimization, Low-Bit Quantization, Inference Compilers
AI Safety & Governance: Statutory Law NLP, Legal AI Alignment, Preference Optimization, Algorithmic Transparency, Sovereign AI Infrastructure