Back to landingProject Archive

All projects.

A focused archive of analytics, campaign strategy, and growth experiments built around measurable business outcomes.

Personal Project - 2026
Customer Churn Prediction & Retention Optimization in Online Food Delivery
End to end Data Project
Dashboard preview

Embedded customer churn food delivery dashboard built in metabase.

Challenge
22.4% Churn Rate
OrderKu faced a 22.4% churn rate and needed to uncover the behavioral and demographic patterns separating loyal customers from at-risk ones, in order to build an actionable retention strategy.
Output
76.5% Churn Recall
tarted with EDA (churn distribution, customer segments, feedback patterns, demographics, geography) to derive customer personas, then benchmarked Logistic Regression vs. Random Forest using 5-Fold Stratified Cross-Validation. The final Random Forest model was optimized via Out-of-Fold threshold tuning with a minimum 40% precision constraint, landing on a 56% decision threshold.
Approach
EDA → ML
The model achieved 76.5% Churn Recall and 61.9% Churn Precision on the holdout test set, with negative feedback emerging as the strongest churn differentiator. Results were deployed via an interactive Streamlit dashboard for customer-level and batch retention prioritization.
View case study
Personal Project - 2026
Bank Customer Churn Prediction System
Machine Learning
Model & Business Signals
20.4%Overall churn rate

Roughly 1 in 5 customers left the bank.

99.8%Model accuracy

Random Forest performance reported in the project.

GitHubSource files

Notebook, Streamlit app, model artifacts, README, and training script are linked from the source repository.

Challenge
Identify high-risk bank customers early enough for retention teams to intervene before churn happens.
Key Result
99.8%
Reported model accuracy using a Random Forest classifier, with complaint behavior emerging as the dominant risk signal.
Approach
Built an end-to-end Python workflow with EDA, feature engineering, model training, saved preprocessing artifacts, and a Streamlit prediction UI.
View case study
Undergraduate Thesis - 2026
Web Based Recruitment System for Agency XYZ
Agile Project Management
User Acceptance Test
86.84%Combined acceptance

"Highly Feasible" — averaged across 15 respondents and 5 usability dimensions.

94.55%Internal users

Agency owner + head of HR — 100% of the back office population (total sampling).

85.54%External users

13 job seekers / public visitors interacting with the front-end website.

Challenge
Agency XYZ ran on a Webflow site that couldn't manage a dynamic database, so recruitment stayed scattered across channels — uncentralized and poorly documented — while the site under-represented the agency to prospective clients.
Key Result
86.84%
Combined User Acceptance Test score across 15 respondents, landing inside the "Highly Feasible" band (80–100%).
Approach
Led delivery with Agile Scrum — product backlog, sprint planning, daily scrum, sprint review, and retrospective — building the system on Laravel and PostgreSQL.
View case study
Looker Studio - 2026
Dashboard KOL
Marketing Dashboard
Live Looker Studio Dashboard

Embedded KOL dashboard built in Looker Studio.

Challenge
Centralize KOL campaign monitoring in a dashboard that is easy to review and share.
Output
Live
Looker Studio report embedded directly into the portfolio as an interactive dashboard.
Approach
Use Looker Studio as the reporting layer for campaign performance tracking, with portfolio access through iframe embed.
View case study
SaaS Product - 2026
Catetin WhatsApp Personal Finance Bot
Full-Stack Development & Automation
Product & Automation Signals
354n8n workflow nodes

Nodes orchestrating the entire conversational experience in production.

6AI agents

LangChain agents on OpenRouter chat models with windowed conversation memory.

12+Integrated services

Supabase, WAHA, Midtrans, Resend, Google Sheets, Upstash, and more.

Challenge
People abandon budgeting apps because logging a transaction takes too many taps, so expenses go unrecorded and the ledger stays incomplete.
Output
WhatsApp
A WhatsApp-first SaaS where users simply chat what they spent or earned, and an AI agent records it into a clean, structured ledger automatically.
Approach
Three connected layers: a Next.js web app, an automated payment-to-activation pipeline, and a 354-node n8n AI engine that does the bookkeeping.
View case study