I use this GitHub profile as a curated portfolio of projects I am building while growing across data science, applied machine learning, analytics, and practical AI/data and IoT systems.

My strongest interest is turning data into useful technical outputs: analysis that explains a problem, models that are evaluated honestly, APIs or telemetry pipelines that make the result usable, and documentation that makes the scope clear.

The portfolio is intentionally broad enough for data science, data analyst/product analyst, AI intern, and applied ML opportunities, while still giving stronger emphasis to projects that connect analysis or modeling with usable workflows.

My current focus is FTM - Fuel Tank Management, an ongoing operational IoT system in active field use within a limited pilot during my practical internship with PT Telkom Infrastruktur Indonesia (TIF/InfraNexia). Registered field devices send live fuel-tank telemetry into a monitored data pipeline that supports operational users with visibility, device management, and support workflows. Additional devices are now being prepared for controlled, needs-based expansion.

I am still developing my depth in this field, so I keep each project clear about its scope, limitations, ownership, and what it is meant to demonstrate.

Ongoing operational IoT system in active field use within a limited pilot during my practical internship with PT Telkom Infrastruktur Indonesia (TIF/InfraNexia). Registered ESP8266 devices connected to ultrasonic sensors send live fuel-tank telemetry through an authenticated Next.js ingestion service into Aiven MySQL, supporting operational monitoring for users and admins, device provisioning, CSV export, Telegram, and helpdesk workflows.

My contribution: project coordination; application architecture; server-side and database work; identity and device workflows; deployment; integration; testing; review; and documentation, in collaboration with field/device/firmware and UI/UX contributors.

Verified pilot snapshot (13-14 July 2026): 3 registered locations, 3 tanks, and 3 physical devices; 9,272 telemetry readings; 2 devices reporting fresh data and 1 requiring follow-up. Live views use the latest per-device snapshot, while 5-minute historical rollups reduce row growth by about 15x versus storing every 20-second reading.

Quality snapshot (16 July 2026): TypeScript typecheck, ESLint, all 160 automated tests, and the Next.js production build passed on the current main branch.

Shows: operational data systems, IoT telemetry ingestion, full-stack architecture, MySQL data lifecycle, authentication and provisioning, field collaboration, CI, and public-safe technical documentation

Stack: Next.js 16 React 19 TypeScript MySQL ESP8266 Telegram Bot API Vercel Vitest

Status and scope: In active operational use within a limited field pilot; not a nationally deployed or fully production-hardened corporate system. Additional devices are being prepared for controlled expansion, while the repository remains ongoing beyond the internship period through security, calibration, alerting, recovery, access control, observability, and operational analytics work.

Links: Live application · Operational status · Architecture

FastAPI-based ML inference service for a cooperative financing decision-support prototype. It loads an XGBoost model artifact, applies the saved preprocessing flow, returns probability/risk information, and keeps the final financing decision under human review.

Shows: ML API design, inference workflow, preprocessing parity, response schema design, model documentation, human-in-the-loop decision support

Stack: Python FastAPI XGBoost scikit-learn Pydantic Model Card

Scope: Prototype using public/Home Credit-style data, not a production credit decisioning system.

Fullstack demo platform that connects a member financing application flow with ML-assisted review. It includes application intake, backend validation, status lookup, admin review queue, scoring, rescoring, and decision workflow.

Shows: fullstack product flow, backend-to-ML API integration, admin/user workflow design, AI as decision support

Stack: React Vite TypeScript Express REST API ML Integration

Scope: Portfolio demo for workflow integration, not a production financial platform.

Telecom-focused data and ML prototype for KPI forecasting and anomaly candidate detection using public anonymized operator time-series data. This is the project that most directly connects my telecom background with applied ML.

Shows: time-series forecasting, residual-based anomaly scoring, canonical data loading, model documentation, telco analytics direction

Stack: Python PostgreSQL FastAPI Docker SARIMA Time Series

Scope: Prototype for portfolio and learning use, not a production NWDAF, IDS, or private operator monitoring platform.

Collaborative filtering book recommender with a modular ML pipeline, KNN-based recommendation logic, Streamlit interface, and Docker support. It is structured beyond a notebook-only workflow.

Shows: recommender system fundamentals, data pipeline structure, model serialization, Streamlit inference app, deployment-aware organization

Stack: Python Pandas scikit-learn KNN Streamlit Docker

Customer segmentation project using the Olist Brazilian E-Commerce Public Dataset. It turns raw relational e-commerce tables into customer-level behavioral features, evaluates K-Means cluster options, and explains the final customer segments in business-friendly language.

Shows: customer analytics, raw data audit, customer-level feature engineering, clustering evaluation, segment profiling, business interpretation

Stack: Python Pandas NumPy scikit-learn K-Means Matplotlib Seaborn

Scope: Individual machine learning assignment using public anonymized e-commerce data, not production customer labels.

Structured SQL certification and interview-readiness portfolio with 57 documented accepted practice solutions across Basic, Intermediate, and Advanced tracks. It includes challenge tracking, concept notes, a SQL style guide, practice logs, and verified HackerRank SQL Basic, Intermediate, and Advanced certificates.

Shows: SQL fluency, joins and aggregation, CTE/subquery reasoning, query documentation, certification discipline, data foundation for ML and analytics work

Stack: SQL MySQL-style Queries CTE Joins Aggregation Documentation

Credentials: SQL Basic · SQL Intermediate · SQL Advanced

Scope: Public-safe practice and certification evidence, not live certification assessment content.

Analytics dashboard project for employee attrition exploration and intervention prioritization. It uses SQL-based preparation and Power BI reporting to move from descriptive analysis toward action-oriented analytics.

Shows: SQL data preparation, BI dashboard storytelling, KPI design, rule-based risk framework, business-facing analytics

Stack: SQL Server Power BI DAX Data Cleaning KPI Design

Scope: The risk score is rule-based prioritization, not a validated predictive ML model.

A combined learning track that supports the applied projects above: math notes, ML derivations, from-scratch implementation, structured AI development assignments, 21-project Python practice with tests/linting, SQL certification practice, database practice, and NLP/RAG exercises.

Shows: mathematical intuition, structured learning discipline, Python/SQL fluency, notebook-based ML practice, testable small-program implementation, and gradual preparation for applied ML work

Related: ML Path to Mastery, ML Hands-On Learning Lab, AI Development B10 Learning Portfolio, Python Projects Learning Portfolio, HackerRank SQL Certification Prep, Linear Regression from Scratch, NLP with HF Transformers, RAG Parameter Exploration

• I prefer projects with a clear problem, data flow, and practical output.
• I document assumptions and limitations so the project does not claim more than it proves.
• I care about clear data assumptions because models are only useful when their inputs, limits, and decisions can be explained.
• I value field feedback, contributor attribution, public-safe documentation, and small reviewable checkpoints.
• I run type, lint, test, and build checks where the project supports them, and I treat operational data and credentials as private by default.
• I use telecom as a strong domain interest while keeping my portfolio open to broader data and ML problems.
• I am gradually improving project structure, model evaluation, API design, deployment, observability, and operational reliability.