Fathur.

AI and Software Engineer; AI @OCBC; Informatics @ITB;

Python
C/C++
PyTorch
TensorFlow
React
Next.js
TypeScript
Rust
Haskell
Go
PostgreSQL
Docker
Node.js
FastAPI
LangChain
Tailwind

I pay attention to how people interact with systems.

Not the features they use. But the moments when they hesitate.

Crowd walking in the city

Movement as a shared language.

Crowds behave like software. Many intentions. One shared space.

People crossing a street in a crowd

I started watching before I started building.

I didn’t start with code.

I started with questions about clarity.

People walking on a subway platform

Patterns repeat when the environment stays the same.

Code became a way to care. For users I may never meet. For teammates I may never see.

Anonymous crowd walking at night

Empathy, without performance.

Clean systems reduce the need for explanation.

Crowd moving through a crossing

Silence is sometimes the most respectful interface.

When something is clear, people feel more confident using it.

Less friction. Less asking. More moving forward.

Blurred people walking on stairs

The interface should not be the obstacle.

I think in sequences.

Inputs. Transitions. Outcomes. And the space between them.

People moving through an urban space

Good work is often just well-managed transitions.

Engineering, to me, is less about solutions and more about transitions.

My work is shaped by observation.

City movement and crowd

I prefer systems that feel invisible.

Something that doesn't ask for attention — but earns it over time.

Crowd crossing in daylight

A calm interface is a form of respect.

This is how I work.

Carefully. With intention. So others can move without thinking about the system beneath them.

Crowd walking through a city at night

No conclusion. Just continuity.

Education

Institut Teknologi Bandung

B.Sc. in Computer Science

2023 - 2027 · Indonesia

Recent Awards

  • [International] 2nd Place - Data Analysis Competition 2025

    HIMASTA ITS · 2025

  • [National] Finalist - BRIN AIDeaNation 2025

    BRIN · 2025

  • [National] Finalist - Data Mining GEMASTIK XVIII

    Kemdiktisaintek · 2025

  • [International] Most Outstanding Student Award

    Matsuo-Iwasawa Lab, University of Tokyo · 2025

  • [National] Bakti BCA Scholarship Awardee

    PT Bank Central Asia Tbk. · 2025

  • [National] Honorable Mention - Data Analytics Dash COMPFEST 16

    COMPFEST 16 · 2024

  • [National] 5th Place - Objective Quest Airnology 3.0

    BEM FTMM Universitas Airlangga · 2024

  • [National] 4th Place - Data Analytics FIND IT! 2024

    DTETI Universitas Gadjah Mada · 2024

  • [National] 3rd Place - DATA SLAYER 1.0

    HMSD IT Telkom Purwokerto · 2024

  • [National] 1st Place - DINAMIK 17 Competitive Programming Contest

    FPMIPA Universitas Pendidikan Indonesia · 2022

What I built

Projects prove I can build. They turn ideas into working systems. Each one solved a problem or answered a question.

OShi-no-Ko: 32-bit Kernel

OShi-no-Ko: 32-bit Kernel

A 32-bit educational operating system built from scratch with paging, scheduling, hardware drivers, system calls, a shell, and an EXT2 filesystem.

Kernel from ScratchDriversEXT2 Filesystem

C · Operating Systems · Paging · Scheduling

Silberschatz

Silberschatz

A mini DBMS project exploring storage internals, query processing, indexing, concurrency control, recovery, and client-server database architecture.

Mini DBMSQuery ProcessingConcurrency Control

Database Systems · Storage Engine · Indexing · Transactions

Pascal-S Compiler

Pascal-S Compiler

A Pascal-S compiler covering lexical, syntax, and semantic analysis with a DFA-based lexer, recursive descent parser, AST generation, and symbol tables.

Lexer to Semantic AnalysisDecorated ASTSymbol Tables

Python · Compiler Design · DFA · Recursive Descent

Raft Protocol

Raft Protocol

An implementation of the Raft consensus algorithm with leader election, log replication, persistent state, and distributed-system failure handling.

Leader ElectionLog ReplicationPersistent State

Java · Distributed Systems · Consensus · Replication

TorchIsAllYouNeed

TorchIsAllYouNeed

A small PyTorch-like deep learning framework built from scratch with autodiff, layers, losses, optimizers, and machine learning experiments.

Autodiff EngineLayer APIML Experiments

Python · NumPy · Autodiff · Neural Networks

Seq2SeqIsAllYouNeed

Seq2SeqIsAllYouNeed

A from-scratch NumPy deep learning project for CNN, RNN, and LSTM inference with image classification and image captioning experiments.

From-scratch InferenceImage ClassificationCaptioning

Python · NumPy · CNN · RNN

Papers

Papers bring me closer to the theoretical side. Articles force clarity. Code demonstrates execution.

Agent-Based SQL Question Answering System Using Small Language Model

Aug 2025

Developed an agent-based Text-to-SQL system using Qwen-2.5-Coder-32B-Instruct with a multi-area architecture, MCP tool binding, and iterative validation loops. Evaluated on a Spider subset with hybrid similarity and SLM semantic equivalence metrics, and analyzed ambiguity/schema challenges with query reformulation strategies.

Text-to-SQLQwen-2.5-CoderAgentic SystemsMCPSpider DatasetSQL

The Importance of Context and Avoiding Overengineering in Sentiment Analysis Problems

June 2025

Explored four different text preprocessing strategies for sentiment analysis of Indonesian text, comparing them with a Gensim baseline. Contributed comprehensive evaluation framework using TF-IDF and SVM classification.

NLPSentiment AnalysisTF-IDFSVMIndonesian Text

Modified Decision Tree Algorithm for Multivariate Nonlinear Data Imputation

January 2025

Proposed custom decision tree model to handle missing values during data splitting for nonlinear multivariate imputation tasks. Achieved notable average RMSE of 4.98 on synthetic educational dataset.

Decision TreesData ImputationMachine LearningMissing Data

Singular Spectrum Analysis for Multivariate Time Series Forecasting

January 2025

Developed forecasting model using SSA on daily food commodity prices in Bandung, Indonesia. Demonstrated SSA's ability to identify trends, seasonal patterns, and noise in non-stationary time series.

Time SeriesSSAForecastingSVDEconomic Data

Writing

Writing is how I express what I think about programming. Hobbies, observations, technical views.

Understanding Major Assignments in the AI Decade

Reflection on how major university assignments can provide new perspectives in the rapidly evolving AI era.

AIEducationUniversityReflection

Depth-First Search and Breadth-First Search

DFS and BFS algorithms are fundamental algorithms, meaning all programmers should at least know about these two algorithms.

AlgorithmsComputer ScienceProgrammingData Structures

Machine Learning: Decision Tree, Information Gain, Entropy

Decision Tree is one of the Machine Learning algorithms widely used for prediction. This article discusses the basic concepts of Decision Trees.

Machine LearningDecision TreesData ScienceAlgorithms

Experience: One Semester Studying Informatics at ITB

Experience and reflection during the first semester studying Informatics Engineering at Institut Teknologi Bandung.

UniversityITBStudent LifeExperience

Work record

Experience shows where I've applied my skills. Each role was about delivering specific results under real constraints.

Jun 2026 - Present

AI/ML Engineer Intern

OCBC Indonesia

Jakarta, Indonesia

OCBC Indonesia (IDX: NISP) is a publicly listed Indonesian banking and financial services company, offering personal, business, corporate, and wealth management banking services as part of the OCBC Group.

  • Work scope is currently confidential.

May 2025 - Feb 2026

Part-time AI and Software Engineer

Fata Organa

Jakarta, Indonesia

Fata Organa is an Indonesia-based AI software company specializing in speech AI and applied machine learning solutions, and serves as an exclusive partner of Empath, a CAC Corporation division from Japan, to develop technologies such as emotional speech analysis, LLM systems, and AI-powered enterprise applications.

  • Built an agentic SQL question-answering system for a Japanese project management platform, enabling natural-language access to internal databases and reducing engineering information lookup time by ~60%; implemented with Python, LangChain, and LLM-based query planning.
  • Automated the generation of Functional and Technical Specification documents using LLM pipelines, cutting manual documentation effort by ~70% and standardizing specification output across 10+ engineering modules.
  • Developed a developer-task recommendation engine leveraging agentic SQL-QA and structured project metadata, improving task allocation accuracy by ~35% and reducing manual task assignment time for project managers by ~50%.
  • Optimized small language model inference using Ollama, vLLM, and model quantization, achieving 2x faster response time and enabling reliable deployment in CPU-only environments (24GB RAM) with <5% accuracy degradation.

Sep 2025 - Oct 2025

AI Engineer

Legali AI

Remote, US

Legali AI is a San Francisco-based legal technology startup building AI-powered tools that help individuals navigate legal processes by organizing evidence, drafting legal documents, and preparing cases through an all-in-one litigation support platform.

  • Architected a multi-agent legal AI workflow that converts natural-language case descriptions into structured legal actions, enabling end-to-end self-service legal assistance for pro se users and reducing manual legal triage time by ~40%.
  • Built a jurisdiction-aware Retrieval-Augmented Generation (RAG) pipeline integrating document ingestion, OCR, hierarchical parsing, PII anonymization, embeddings, and region-filtered semantic retrieval, improving legal citation relevance by ~30% across multiple U.S. jurisdictions.
  • Developed an AI-driven legal document generation system that allows users to draft, revise, and export court-ready filings, reducing document preparation time from hours to minutes through template-guided prompting and legal clause retrieval.
  • Engineered scalable AI orchestration services with asynchronous processing, conversational memory, and source-traceable outputs, supporting 1,000+ document queries/day while maintaining reliable and explainable legal reasoning pipelines.

Pro bono work

Unpaid work that sharpens skills and gives back. These roles tested my ability to design problems and validate solutions.

Apr 2026 - Present

Head of Tech Issues and Exploration

HMIF ITB

Tech Issues and Exploration is a department consisting of Competition and Community, Career Development, and Technology Development.

Coordinated strategic alignment across Competition and Community, Career Development, and Technology Development to ensure each division's objectives, programs, and monthly evaluations run according to the department roadmap

Handled strategic negotiation between HMIF ITB and DITSP to reduce hosting platform billing costs from IDR 2.9 million per year to zero, improving operational efficiency without reducing infrastructure support

Supported the department's role in maintaining technology relevance by building internal tools such as WhatsApp bots and LINE bots to support other divisions, streamline workflows, and improve organizational productivity

May 2025 - Mar 2026

Head of Competition and Community

HMIF ITB

Competition and Community is a division under HMIF ITB that supports students in developing their passion, preparation, and achievement in technology competitions.

Managed strategic planning for GEMASTIK preparation, including coordination, learning direction, and competition-readiness initiatives for HMIF ITB members

Managed funding support for competition finalists to help members participate in final-stage competitions more effectively

Built and maintained competition archives, including learning resources, past materials, and documentation, so future cohorts can access structured references for Data Science competition preparation

Nov 2024 - Present

Problem Setter

ARKAVIDIA 9.0

Developed and structured data science and machine learning challenges for Datavidia, ensuring a diverse range of problem difficulties and real-world applications. Also conducted extensive data collection and feasibility testing on problems.

Developed and structured data science and machine learning challenges for Datavidia

Conducted extensive data collection using Selenium for web scraping and Pandas for preprocessing

Performed feasibility testing on problems by evaluating model performance with SHAP, RNN, Prophet, and statsmodels

Collaborated with other problem setters to validate problem accuracy, fairness, and adherence to competition standards

May 2024 - Jul 2024

Problem Setter

IMPACT ITB 4.0

Designed and formulated high-quality algorithmic and mathematical problems for the competition. Created problem sets across multiple domains and developed Higher-Order Thinking Skills (HOTS) problems to test participants' deep problem-solving skills.

Designed and formulated high-quality algorithmic and mathematical problems for the competition

Created problem sets across multiple domains, including Mathematics, Physics, and Informatics

Developed Higher-Order Thinking Skills (HOTS) problems covering advanced topics such as Dynamic Programming and Graph Theory

Reviewed and tested problems to guarantee correctness, fairness, and feasibility under competition constraints

Figures

A way to remember what I've done. Visual memory of where I've been and what happened.

Selected figure 01

A pause within a shared moment.

Selected figure 02

Paths taken between obligations.

Selected figure 03

Tools arranged for quiet work.

Selected figure 04

Routine measured in steady stride.

Selected figure 05

Motion without intent to be seen.

Selected figure 06

Time held before moving on.

Selected figure 07

A moment passing without notice.

Selected figure 08

Scale felt through distance.

Selected figure 09

Presence caught in transit.

Selected figure 10

A frame that remains.

When the work aligns.

My work sits between software, data, and systems.

When a project needs clarity, structure, and long-term attention, I am available.

If the work makes sense, we can talk.

Muhammad Fathur Rizky (Fathur) | AI and Software Engineer; AI @OCBC; Informatics @ITB; | Fathurwithyou