Hello there,
I'm Richard. Welcome to my page!
You can find here small, practical, problem-solving case studies covering some details about automation, automation tools, AI, guides, webpages, and more. For any question, feel free to contact me.
See you.
Deterministic life path simulator. Type any scenario — AI generates a decision tree with real probabilities from 350K+ data points. 36 sacred roots (Bible + Quran) as gate evaluators. Features: Crash Test (compare 3-5 paths), Reverse Engineering (trace outcomes back), 3D walk-through, photo/voice input, 28 templates.
Next.js 16 · React Flow · Claude Haiku · Supabase pgvector (66K embeddings) · Three.js · Dagre · Web Audio
Live (private deploy) · Watch Demo · Technical Details · Decision Log
Decision tree from "I want to move to Japan and teach English." — 14 branching nodes, real probabilities from 350K+ data points, 36 sacred roots evaluation. Generation: 3.2s.
350K+ data points · 114 JSON files · 7 live APIs · 28 templates · Cross-device PIN sync
Problem: Life decisions are high-stakes but people decide on gut feeling.
Decision: Deterministic trees with RAG (66K embeddings in pgvector) + Claude Haiku + sacred texts as gates.
Trade-off: Complex architecture vs pure-LLM.
Result: 350K+ data points, 28 templates, Crash Test — all at $0/mo.
AI-powered document automation: upload a PDF, extract case data, generate completed legal motions in seconds.
Python · Claude AI · pdfplumber · Streamlit
Watch Demo · Technical Details · Decision Log
Uploaded 12-page case PDF → extracted 23 data fields (parties, dates, claims, jurisdiction) → generated Motion to Dismiss in 8 seconds, court-filing ready.
PDF parsing · Structured data extraction · Court-ready document generation · Zero-friction UI
Problem: Hours manually extracting data from case PDFs.
Decision: pdfplumber for extraction + Claude for parsing + Streamlit for UI.
Trade-off: Cloud API dependency vs offline.
Result: PDF to motion in under 10 seconds, 90%+ format coverage.
Machine-learning image recognition app that finds your photos by keyword. After the initial index and with your preferred model (I implemented more than one), you can search images by tags, text (OCR), visual similarity, date, or folder filters. It is fully offline and private.
Python · CLIP · BLIP · EasyOCR · FAISS · PyQt6 · SQLite
Watch Demo · Technical Details · Decision Log

Multi-model indexing · Offline-first · Tag/OCR/similarity search · Cross-platform desktop GUI
Problem: Photo libraries aren't searchable by meaning.
Decision: Offline indexing pipeline with multiple model options.
Trade-off: Slower first scan + larger index.
Result: Private local search by tags/OCR/similarity.
Automated price-drop detection system scraping Indonesian marketplaces, filtering deals, and publishing alerts via Telegram.
Google Apps Script · Playwright · Telegram Bot API
Live · Watch Demo · Technical Details · Decision Log
Detected 42% price drop on Tokopedia → filtered through deal criteria → Telegram alert in 90 seconds.
Google Apps Script · Playwright · Telegram Bot API
Multi-marketplace scraping · Price history tracking · Configurable deal filters · Instant Telegram alerts
Problem: Deals disappear fast, manual monitoring impractical.
Decision: Google Apps Script (free cron) + Playwright + Telegram Bot API.
Trade-off: GAS 6min limit; Playwright heavier but necessary.
Result: Scan → filter → alert in under 90s, $0/mo.
Execution notes documenting some step-by-step I went through and system implementation.
Step-by-step technical guides from real projects.
Side explorations — tools and prototypes I built to learn, test ideas, or solve personal problems.
Cafepedia.id — Cafe discovery platform, 787K+ places indexed (Indonesia)
Knowledge Base World — 3D knowledge graph, 180+ tech keywords visualized
Kalkulator Cicilan — Installment calculator for Indonesian consumers
Alatwarung — Equipment catalog for street vendors