From Teaching Assistant to Server Architect: Tackling Flood Risks and Next.js

Welcome back to another weekly update from my internship at the KMUTT Geospatial Engineering and Innovation Center (KGEO). This week (October 5th to 9th) brought a massive shift in my responsibilities, bouncing between web development, teaching, tackling a brand-new machine learning challenge, and engineering server infrastructure. 

Wrapping Up AGB and Stepping into the Classroom

I kicked off the week by officially submitting my finalized academic report on the Mangrove AGB estimation. With that major milestone complete, I immediately pivoted to web development. I began migrating the frontend of the WefGIS website (a collaborative platform between Undiksha and KMUTT) from basic HTML to a modern Next.js architecture. You can access the website in https://wefgis.com/


Monday also brought a completely new experience: I stepped in as a Teaching Assistant for Mr. Agus's Data Mining class. Assisting an international class for the first time was incredibly fun, exciting, and a fantastic way to practice my communication skills in an academic setting.


Deploying WefGIS and A New Prediction Challenge

By Tuesday, the Next.js migration for WefGIS was complete. I also took the time to rebuild and deploy our past SIPETANI project alongside the WefGIS application, ensuring everything ran flawlessly on the TRPL Study Program's Proxmox server back home.


Just as I cleared my desk, Pi Nalin handed me an exciting new project: flood risk prediction. The twist? It requires cross-provincial modeling. I need to train the machine learning model using a dataset from Nakhon Pathom and deploy it to predict flood risks in Sukhothai province. I spent Wednesday heavily planning this workflow while simultaneously finalizing the draft of my Bali Drought Analysis paper.

Troubleshooting Cloudflare and Executing the Model

Thursday was a true test of problem-solving. I started the day troubleshooting the 5TB KGEO storage server I built last week. Users were hitting a wall when trying to upload gigantic files (over 2 GB). After some digging, I identified the root cause as a Cloudflare proxy limit and began drafting a configuration repair plan for the next day.

Later, I finally executed the cross-provincial flood risk model. The initial accuracy was surprisingly good, but it definitely requires further refinement. The process taught me a valuable lesson: looking at a dataset from a completely different perspective is often the key to unlocking better model performance and finding new areas for upgrade.

Engineering a Headless Compute Server

I ended the week with a hardcore IT infrastructure project. I decided to turn my primary KGEO workstation into a shared computing server so the rest of the team could utilize its hardware for heavy model training. The ultimate constraint? I had to configure it purely via SSH to run headless, without formatting the drive or disrupting the existing file system.

While the initial stress tests revealed some bottlenecks in connection stability and remote resource allocation, the proof of concept works. Figuring out how to maximize hardware utility and securely share resources without resetting the PC was a fantastic technical puzzle to close out an incredibly productive week.

Comments