Projects
A showcase of systems engineering, backend services, and automation pipelines built to solve real-world problems.
Backend & Full Stack
WhatsApp Automation & AI CRM
Automated CRM system for offline businesses to manage bookings and upsell customers via WhatsApp integration.
- Built a backend webhook handler (FastAPI) to receive WhatsApp chats via WAHA.
- Integrated OpenAI API to analyze conversation context and automatically extract booking data.
- Used APScheduler to send automated appointment reminders 1 day before visits.
- Implemented customer retention triggers that automatically send upsell messages after 1 month.
FastAPIWAHAOpenAISupabaseAPScheduler
CCTV Intelligence & Report Daemon
System that processes hotel CCTV footage to generate voice-only transcriptions, AI summaries, and email daily manager reports.
- Designed a custom Linux daemon that receives CCTV video uploads via FTP.
- Implemented Voice Activity Detection (VAD) via FFmpeg to crop silent parts, reducing OpenAI token cost.
- Generated daily transcriptions and AI summaries of human speech, delivering them to managers via Zoho SMTP.
- Developed a lightweight Flask dashboard and automated a script to auto-cleanup data files after 3 months.
FlaskOpenAI APIFFmpegVAD (Voice Activity Detection)Zoho SMTP
Hardware-Integrated Parking MVP
Web-based parking management system prototype that directly interfaces with gate hardware, cameras, and ticket printers.
- Developed a web-based replacement for outdated desktop-dependent parking software.
- Integrated gate controllers using physical I/O signal protocols to handle entry/exit gate operations.
- Connected local cameras using the ONVIF protocol for automated vehicle capture.
- Designed print routines for instant thermal ticket generation.
FlaskMySQLONVIF ProtocolI/O Gate ControllersThermal Printing
QA Automation & Web Scraping
QA Automation for Solar Platform
Comprehensive E2E and unit testing pipeline for a Canadian solar panel layout planning platform running in GitHub Actions CI/CD.
- Wrote robust smoke tests, API validation tests, and end-to-end workflow scenarios.
- Solved complex testing challenges involving nested Google Aerial Maps iframes.
- Integrated test execution directly into the GitHub Actions CI/CD pipeline.
PlaywrightTypeScriptGitHub Actions
Cost-Optimized High-Volume Scraper
High-throughput data extraction pipeline gathering content from podcasts, blogs, and Reddit forums for RAG database ingestion.
- Processed over 7,000 YouTube podcast transcripts and cleaned text using Gemini 2.5 Flash.
- Implemented LLM prompt caching that reduced API token expenses by 50% ($100 total cost).
- Used Trafilatura for metadata extraction and SeleniumBase + Capsolver to bypass Cloudflare Turnstile.
- Built scraper fallbacks using raw Scrapy spiders for standard server-rendered HTML sites.
PlaywrightScrapyGemini 2.5 FlashSeleniumBaseTrafilaturaCapsolver API