Portfolio Archive

An end-to-end AI-driven predictive pipeline and interactive web dashboard designed to forecast hourly electricity demand for the Tokyo Electric Power Company (TEPCO). By integrating multi-year time-series data, weather APIs, and Deep Learning models (PyTorch LSTM), the system provides high-accuracy forecasts (MAE ~150 MW) and real-time peak-shaving simulations, empowering energy stakeholders to proactively mitigate outage risks and significantly reduce emergency power procurement costs.

1. Context & Business Challenges

• Grid Stability & Load Balancing: Managing large-scale energy demands requires precise hourly load balancing in the Kanto region to ensure grid stability.
• Weather-Driven Demand Spikes: Extreme weather anomalies (temperatures >28°C) create high uncertainty in electricity demand surges.
• Operational & Financial Risks: Inaccurate forecasting leads to a high risk of power outages and massive budget overruns caused by sudden, expensive emergency power purchases on the spot market.

2. Technical Solution & Role (Data Scientist / Full-Stack Developer / UI/UX Designer) Designed and engineered a scalable, end-to-end predictive infrastructure and an interactive web application:

• Automated ETL Pipeline: Built robust ETL scripts to clean and process massive multi-year time-series datasets, handle Japanese data encoding parsing, and integrate secondary weather attributes via external APIs.
• Hybrid AI Modeling (PyTorch & Prophet): Developed a predictive architecture comparing the reliability of Probabilistic Additive Models (Facebook Prophet) against Deep Learning Neural Networks (PyTorch LSTM) to effectively capture anomaly trends during extreme weather.
• Interactive Web Dashboard (Streamlit, Pandas, Matplotlib): Transformed complex analytical logic into a seamless, high-performance UI/UX. The dashboard features a memory-efficient, real-time “Peak Shaving Simulator” equipped with visually actionable elements like interactive temperature heatmaps and demand duration curves.

3. Business Impact & Insights

• High-Precision Forecasting: The implemented PyTorch LSTM model achieved highly accurate and consistent predictions, successfully narrowing the error margin to a Mean Absolute Error (MAE) of ~150 MW.
• Cost Reduction & Efficiency: Delivered a critical decision matrix to corporate management, enabling them to simulate peak-shaving strategies and avoid millions of Megawatts of inefficient, high-cost power procurement.
• Proactive Risk Mitigation: Successfully translated complex mathematical data into precise visual insights, empowering executives to proactively identify grid stress and mitigate potential outages before they occur.

An end-to-end data science and analytics project designed to transition Olist e-commerce from assumption-based to data-driven decision-making. By integrating complex relational databases, applying K-Means Clustering for RFM segmentation, and utilizing Facebook Prophet for time-series forecasting, the project delivered a centralized interactive dashboard that optimizes targeted marketing, proactive inventory planning, and logistics evaluation.

1. Context & Business Challenges

• Lack of Data Visibility: The company needed a centralized data overview to shift towards a data-driven strategy.
• Customer Behavior Mapping: Faced difficulties in mapping the behavior of millions of users to create personalized marketing campaigns.
• Demand Uncertainty: The inability to accurately predict market demand fluctuations led to high risks of overstocking or stockouts.
• Logistics & Customer Satisfaction: Needed to identify the root causes of high volumes of poor customer reviews and connect them with logistical performance metrics.

2. Technical Solution & Role (Lead Data Scientist) Led the entire single data processing lifecycle (End-to-End Data Science), culminating in a unified interactive analytical dashboard:

• Data Engineering & Integration: Executed data cleaning and complex feature engineering by integrating 9 high-value relational tables.
• Customer Segmentation (K-Means Clustering): Applied logical mathematical approaches to execute an RFM (Recency, Frequency, Monetary) customer segmentation model.
• Time-Series Forecasting (Facebook Prophet): Designed and trained an advanced analytical model to handle fluctuating demand and heavy seasonality (e.g., Black Friday effects), optimized further by incorporating Brazilian holiday data.

3. Business Impact & Insights

• Cost Savings & Optimal Retention: Successfully divided customers into 4 essential segments (such as “Champions” and “Hibernating”). This allowed the marketing team to deploy highly targeted, cost-efficient win-back campaigns instead of distributing random, budget-draining promotions.
• Financial & Inventory Planning: The high-accuracy forecasting revealed a stable upward trend in the “Health & Beauty” category. This provided robust analytical justification for inventory management to increase buffer stock by 10-15%, ensuring the company captured the next quarter’s profit momentum.
• Logistics Mitigation: Statistically and causally validated the negative correlation between regional delivery inefficiencies and dropping review scores. This provided executive directors with concrete, unarguable evidence needed to review and overhaul their logistics Service Level Agreements (SLAs) and SOPs.

A standalone desktop application designed to modernize the complex financial and administrative workflows of an educational institution. Engineered as an offline-first solution using web technologies, the system successfully digitizes bureaucratic data entry, minimizes human error, and provides real-time, highly reliable financial tracking without dependency on internet connectivity.

1. Context & Business Challenges

• Complex Administration: The institution handles highly specific administrative workflows, including dormitory (hujroh) management, daily institutional expense tracking, and thousands of student transactions with varied fee structures (e.g., meal plans, laundry, building donations).
• High Risk of Human Error: Manual bookkeeping processes resulted in a high risk of calculation errors and made financial audits extremely difficult.
• Infrastructure Limitations: Unstable internet access within the institution required a robust system capable of running fully offline without compromising data security or accuracy.

2. Technical Solution & Role (Lead Full-Stack Developer) Designed and developed a comprehensive desktop application to overhaul the institution’s administrative pipeline. The system features a financial analytics dashboard, student and dormitory management modules, daily transaction tracking, dynamic tariff settings, and a digital receipt generator. Built for high-reliability offline use with a lightweight architecture:

• Desktop Framework: Electron.js.
• Frontend Logic & Styling: Vanilla JavaScript & Tailwind CSS to ensure a clean UI/UX, rapid navigation, and responsive screen components.
• Database & Storage: Local Database Engine using SQLite (via sql.js) embedded directly into the application. Implemented a .db Export/Import feature for independent data backup management and base64 conversion for the receipt file upload system.

3. Business Impact & Insights

• Minimized Human Error: The system automated payment calculations, arrears management, and historical tariff synchronization, ensuring 100% accuracy in financial tracking.
• Operational Efficiency: Automated cash-flow logging and structured digital receipt archiving drastically accelerated the institution’s financial audit processes, reducing bureaucratic data entry to just a few clicks.
• Institutional-Level Reliability: By running the application entirely in a local desktop environment (serverless), administrative staff are completely freed from network latency issues, guaranteeing uninterrupted, real-time service for student payments and dormitory operations.

A highly precise, mini-ERP web application designed to automate post-employment benefit calculations under the PSAK 219 accounting standard and Indonesian Labor Law (UU Cipta Kerja). By engineering custom mathematical logic to overcome standard floating-point limitations, the system perfectly replicates Microsoft Excel’s calculation engine, delivering 100% audit-ready precision and reducing corporate valuation time from days to minutes.

1. Context & Business Challenges

• Regulatory Compliance: Companies struggle to independently calculate complex post-employment benefits under the strict guidelines of PSAK 219 and PP No. 35 Tahun 2021.
• The Precision Bottleneck: Public Accounting Firm (KAP) auditors strictly demand that the final actuarial valuation matches actuary-standard Microsoft Excel calculations down to the exact last digit.
• Floating-Point Limitations: Standard modern web applications frequently fail audit tests due to inherent calculation discrepancies between standard programming floating-point arithmetic and Excel’s native calculation engine.

2. Technical Solution & Role (Lead Full-Stack Developer) Designed and developed a specialized mini-ERP system focused on HR data and actuarial parameters, automating the Projected Unit Credit (PUC) formula:

• System Architecture: Built an end-to-end platform featuring a participant census database, salary increment projectors, and a flexible Rules Engine to accommodate various employee termination scenarios.
• Mathematical Precision Engineering: Integrated the Decimal.js library into a custom module to execute arbitrary-precision, base-10 arithmetic, entirely bypassing standard floating-point distortion. Developed custom backend algorithms to specifically mirror Microsoft Excel’s Symmetric Arithmetic Rounding.
• Modern Tech Stack: Frontend developed with Next.js (React), TypeScript, Tailwind CSS, and Radix UI for a clean, corporate-standard interface. Backend powered by Node.js (API Routes) and NextAuth.js. Data managed via PostgreSQL (specifically configured to store exact numeric/decimal values rather than approximations) using Prisma ORM, deployed via CI/CD pipelines.

3. Business Impact & Insights

• Audit-Ready & 100% Accurate: The application successfully aligned billions in Defined Benefit Obligation (DBO) calculations to achieve immediate audit validity. It requires zero manual reconciliation from KAP, completely eliminating financial discrepancies caused by human error.
• Operational Efficiency: Drastically accelerated the valuation process for entire corporate populations, cutting down processing time from several days of manual Excel labor to just a few minutes.
• Strategic Flexibility: Empowered HR departments to perform “on-the-fly” financial liability calibrations and severance scenarios for specific individuals (updating PVBO). Executives can now instantly generate decision-making metrics for severance funds without needing to recalculate the entire company’s population data.

Equilibrium-X is an AI-driven pricing agent utilizing Causal Inference and Game Theory to navigate market volatility and logistics disruptions. It provides executives with a compliance-safe, real-time interactive dashboard to optimize pricing strategies, successfully projecting up to a 43% increase in net margin while mitigating the risks of price wars and algorithmic collusion.

1. Context & Business Challenges

• Market Volatility: Facing operational cost crises due to logistics disruptions (“The 2024 Problem”).
• Traditional ML Limitations: Existing algorithms only read historical data correlations and fail to distinguish cause-and-effect. They cannot reliably answer strategic questions regarding the risk of price wars if the company raises selling prices.
• Regulatory Constraints: Management requires a robust automated system that strictly complies with anti-monopoly regulations and the ban on Algorithmic Cartels (Tacit Collusion – AB 325 Regulation).

2. Technical Solution & Role (Lead Data Scientist / AI Developer) Designed the end-to-end architecture of the decision engine and War Room Dashboard using an advanced scientific approach:

• Causal AI Engine (Python, Microsoft DoWhy): Built causal inference algorithms to discover true price elasticity by isolating confounding variables (e.g., holiday seasons vs. price promotions). The engine’s reliability is validated using Placebo/Dummy Tests to ensure zero statistical hallucinations.
• Game Theory Algorithm (Python, Nashpy): Applied competitive market matrix calculations to find the optimal and secure pricing point (Nash Equilibrium) for the company, even when facing aggressive pricing from competitors.
• Strategic Dashboard & Governance (Streamlit, Pandas, Plotly): Designed an interactive UI/UX that projects margin vs. volume scenarios to executive levels. The system is equipped with real-time legal detection (“Algorithmic Police”) that alerts against mark-up tactics exceeding fair competition limits (mark-ups > 20% without validated reasons).

3. Business Impact & Insights

• Data-Driven Insights: The Strategic Simulation module revealed counter-intuitive findings: during a logistics crisis surge, holding prices to maintain market share will actually destroy profit.
• Corporate Impact: The system drove a crucial shift in business strategy, recommending a “Premium” pricing approach to facilitate volume shedding. Using the Nash Equilibrium recommendations, the company projected a Net Margin increase of up to 43%, despite a temporary drop in total volume.
• Legal Compliance: The Governance module successfully aligned financial decisions to be 100% safe and compliant with state anti-collusion regulations, protecting the corporation from legal risks and fines.

A multi-tenant B2B SaaS platform engineered for Hajj and Umrah travel agencies to digitize and automate the evaluation of pilgrims’ ritual (manasik) readiness. Designed with an elderly-friendly interface and a robust Supabase backend, the system eliminates administrative bureaucracy, provides real-time analytics, and elevates the agency’s professional service standards.

1. Context & Business Challenges

• Operational Inefficiency: Travel agencies faced layered operational challenges in evaluating pilgrims’ understanding of rituals. Reliance on manual simulations and paper-based tests resulted in heavy administrative bureaucracy.
• Lack of Visibility: Management lacked a comprehensive, centralized overview of each individual pilgrim’s readiness prior to departure.
• Demographic Hurdle: The primary user base is dominated by elderly individuals who frequently struggle with standard, complex e-learning applications, requiring a highly specialized UX approach.

2. Technical Solution & Role (System Architect & Full-Stack Developer) Designed and developed the entire multi-tenant B2B SaaS platform from the ground up, catering to two primary user segments:

• B2B Management Dashboard: Built a Live Analytics Dashboard featuring dynamic quiz and material management, auto-grading, WhatsApp distribution integration, and real-time performance report generation (CSV & PDF).
• B2C Pilgrim Portal (Elderly-Friendly UI): Engineered a frictionless evaluation portal specifically designed for senior users, implementing large typography, high-contrast elements, and absolute navigation to ensure independent use with minimal technical friction.
• Advanced Architecture & Security: Utilized Next.js (App Router), React 19, TypeScript, Tailwind CSS v4, Framer Motion, and Recharts. Architected the backend using Supabase (Auth, PostgreSQL), implementing Multi-Tenant Row Level Security (RLS) to guarantee strict, independent data isolation between different travel agencies.

3. Business Impact & Insights

• Operational Efficiency & Accuracy: Drastically reduced physical data bureaucracy and manual recapitulation. The auto-grading system guarantees 100% assessment accuracy, effectively eliminating administrative human error.
• Smart Business Insights: The trend analytics dashboard empowers agencies to identify specific ritual topics (e.g., logistics, health, history) that a pilgrim batch struggles with, allowing for highly targeted and efficient pre-departure briefings.
• Elevated Service Quality & Reliability: The system’s robust real-time processing handles massive concurrent access flawlessly. Its inclusive, elderly-friendly design ensures high completion rates, directly helping the travel agency modernize and elevate its professional service standards.

A specialized B2B digital platform developed for imbalankerja.id using a WordPress-based architecture. Beyond technical development, this project involves a comprehensive content strategy where I serve as the primary technical writer. The platform streamlines PSAK 219 actuarial calculations and serves as a high-authority educational hub for corporate HR and Finance professionals.

1. Context & Business Challenges

• Regulatory Complexity: Most companies lack internal actuaries to calculate post-employment benefits according to PSAK 219 (formerly PSAK 24) standards, leading to high risks of financial misstatement.
• Information Gap: There is a significant need for clear, accurate, and up-to-date educational material regarding the intersection of Labor Law (UU Cipta Kerja) and accounting standards.
• Audit Obstacles: Inaccurate or delayed actuarial calculations frequently obstruct external audit processes, causing operational delays for management.

2. Technical Solution & Role (Full-Stack Developer & Content Strategist) Responsible for the end-to-end design of the B2B web infrastructure and the execution of its technical content strategy:

• Custom CMS & Information Architecture: Implemented a scalable WordPress architecture to manage complex actuarial service listings alongside a high-volume educational blog.
• Technical Content Production: Authored a comprehensive library of technical articles focusing on PSAK 219 compliance, severance calculations, and HR-tech integration, translating complex regulations into actionable business insights.
• Operational Automation: Engineered interactive form systems to automate the quotation request workflow, effectively bridging the gap between digital readers and service procurement.
• SEO & Inbound Marketing: Optimized all technical content for search engines to attract organic corporate traffic, specifically targeting Finance and HR decision-makers.

3. Business Impact & Insights

• Authority Building: By consistently producing high-quality technical articles, the platform established itself as a “Thought Leader” in the Indonesian actuarial space, significantly increasing brand trust and client acquisition.
• Workflow Acceleration: The structured platform reduced the time required to generate audit-ready reports, while the educational content served as a pre-sales tool that addressed common client inquiries automatically.
• Reduced Friction: The combination of intuitive UX and clear technical documentation minimized data entry errors from clients and cut down the administrative bureaucracy previously associated with manual consultations.

A large-scale recommendation engine designed to combat “choice paralysis” and enhance user retention by processing over 10 million MyAnimeList interactions. This hybrid system integrates Collaborative Filtering (SVD) and Content-Based Filtering (TF-IDF) into a memory-efficient pipeline using PyArrow and Parquet, deployed as a scalable FastAPI microservice with an interactive Streamlit dashboard.

1. Context & Business Challenges

• User Retention & Choice Paralysis: Digital entertainment platforms struggle with retention due to information overload, where users often spend more time searching for content than actually consuming it.
• Limitations of Generic Discovery: Standard “Top 10” or popular lists are insufficient for maintaining long-term interest; users require hyper-personalized suggestions to stay engaged.
• Information Overload: Navigating vast catalogs without tailored guidance leads to user disorientation and platform abandonment.

2. Technical Solution & Role (End-to-End Data Scientist) Engineered a comprehensive hybrid recommendation architecture from data ingestion to deployment:

• Hybrid Recommendation Logic: Developed a dual-engine approach combining Collaborative Filtering (SVD) to capture latent user preferences and Content-Based Filtering (TF-IDF & Cosine Similarity) to analyze metadata such as genres, studios, and synopses.
• High-Performance Data Pipeline: Utilized PyArrow and Parquet formats to ensure the efficient processing of 10 million+ interaction records, allowing for smooth computation on standard hardware.
• Microservice Deployment: Architected the system as a FastAPI-based REST API microservice and built an interactive Streamlit dashboard for real-time predictions and metadata visualization.

3. Business Impact & Insights

• Personalized Accuracy & Content Diversity: The hybrid model (60% Collaborative, 40% Content-based) effectively prevents “echo chambers” by balancing popular titles with unique, niche discoveries that match specific user tastes.
• Cold Start Resolution: Implemented an intelligent fallback mechanism to handle new users or items with no historical data, ensuring an engaging experience from the very first interaction.
• Operational Efficiency: The memory-efficient architecture significantly minimized latency, resulting in a faster, more responsive user interface (UX) during real-time searches.

A high-performance B2C travel platform developed for PT Azzura Tour International using a customized WordPress ecosystem. The project focused on transforming complex pilgrimage information into a transparent, accessible digital catalog, featuring seamless WhatsApp integration and optimized UI/UX to drive digital conversions and strengthen brand trust.

1. Context & Business Challenges

• Information Transparency: The agency required a digital platform to present comprehensive details for Umrah, Hajj, and Halal Tourism packages transparently.
• Accessibility for Diverse Demographics: The primary challenge was creating a digital catalog easily navigable by pilgrims across various age groups.
• Educational Integration: The client needed a centralized hub that combined travel listings with educational content (Manasik rituals and travel tips) to build a trustworthy brand authority.

2. Technical Solution & Role (Web Developer) Responsible for the end-to-end design and configuration of a customized WordPress (CMS) ecosystem:

• Custom Architecture: Developed a custom data structure for travel package management, ensuring administrative ease for real-time content updates.
• Seamless Communication Flow: Integrated WhatsApp API directly into the reservation system to facilitate immediate interaction between pilgrims and the agency.
• Optimization & Security: Configured essential plugins for high performance, robust security, and advanced SEO to ensure maximum visibility and a smooth mobile/desktop experience.
• Tech Stack: WordPress, PHP, MySQL, JavaScript, Elementor Pro, CSS3.

3. Business Impact & Insights

• Reduced Operational Bottlenecks: The structured presentation of package details (airlines, hotels, and costs) significantly reduced repetitive inquiries for the administrative team, allowing them to focus on high-value tasks.
• Increased Brand Trust: Transparent and well-organized information directly improved customer confidence, positioning the agency as a professional and modern service provider.
• Conversion Growth: By eliminating technical friction and optimizing the customer journey—from reading travel tips to selecting a package—the platform successfully increased lead generation through digital channels.

A high-traffic digital publication platform where I serve as the Lead Web Manager and Content Strategist. The project focuses on maintaining consistent organic growth and operational stability through a highly optimized, SEO-friendly web architecture. By streamlining the daily publication workflow and enforcing strict technical SEO standards, the platform delivers real-time information to a B2C audience with maximum performance and minimal latency.

1. Context & Business Challenges

• Traffic Volatility & Visibility: As a digital news/publication portal, Keunal.id required consistent visibility on search engines to sustain organic traffic growth.
• Operational Bottlenecks: The need for daily, high-frequency updates demanded a content management system that eliminated technical bureaucracy, allowing for rapid article deployment.
• User Engagement: High bounce rates and slow indexing were critical risks that needed to be addressed through a lightweight and responsive web architecture.

2. Technical Solution & Role (Web Manager & SEO Strategist) Responsible for the full operational lifecycle of the web ecosystem, from technical architecture to strategic content execution:

• SEO-Driven Architecture: Engineered the site structure to meet advanced search engine standards, focusing on Technical SEO elements such as page load speed, metadata schemas, and automated indexation flows.
• Content Workflow Optimization: Configured a custom CMS framework that drastically simplified the “draft-to-publish” pipeline, ensuring high-frequency updates without technical friction.
• Data-Driven Strategy: Utilized web analytics and search console data to perform keyword research, informing both the technical layout and the strategic direction of daily articles.
• Responsive Maintenance: Ensured 100% cross-device responsiveness and system reliability to handle real-time traffic surges during trending news cycles.
• Tech Stack & Tools: Web CMS, HTML/CSS, Google Search Console, Google Analytics, SEO Management Suites.

3. Business Impact & Insights

• Organic Growth & Visibility: Strategic SEO integration resulted in a significant surge in organic search visibility and reader engagement metrics, strengthening the platform’s domain authority.
• Operational Efficiency: The streamlined publication workflow minimized human error and reduced the time required for deployment, allowing the editorial team to focus on content quality rather than technical hurdles.
• Performance-Driven UX: A lightweight architecture ensured a frictionless reading experience with zero latency, leading to a noticeable decrease in bounce rates and increased session durations.
• Scalability: The robust system maintenance ensures the platform can scale seamlessly as the article database grows, maintaining peak performance regardless of traffic volume.