Case study· Jul 2023 – Sep 2023

WOWS — Automating 9 Daily Reports for BBMKG's Early-Warning Division

Solo internship build at BBMKG Wilayah III Denpasar's MEWS division. A Python + Streamlit web app that replaces the daily handwriting of nine .docx reports with a one-click form, delivered in nine weeks.

9 daily reportsDocuments automated

9 weeksBuild duration

Solo (intern)Team size

Python · StreamlitStack

StreamlitPythonWeb AutomationInternship

Problem

BBMKG Wilayah III Denpasar is one of Indonesia's regional meteorological agencies. Inside it, the MEWS division (Meteorology Early Warning System) produces nine mandatory .docx reports every shift — weather summaries, equipment logs, observation handovers — each filled in manually from data scattered across observation tools, spreadsheets, and shift notes.

Two problems compounded:

The same fields were being typed by hand, every shift, every day. Predictable transcription errors, hours of duplicated effort.
There was no live view of which staff member was on duty. Schedules were managed by a different division, so the MEWS lead had no quick way to see who was currently in the room.

I joined as a single Universitas Udayana intern in July 2023. The brief: fix both, in the time I had on site.

Data

Nothing fancy on the data side — that was the point. The MEWS team handed me:

The nine .docx templates that needed to be filled out daily.
The fields each report expected (shift, observer, observation values, supervisor notes, miscellaneous).
Their existing on-shift roster format, used to decide who would appear on the monitoring board.

The work was understanding the reports and the workflow, not collecting external data.

Approach

The internship ran from 10 July 2023 to 10 September 2023, with the final presentation on 11 September 2023 — about nine weeks end-to-end. I structured it in five stages:

Requirements & scoping. Sat with MEWS staff to walk through the actual document workflow, not the documented one. Confirmed which nine reports were truly mandatory and what fields each expected.
Tool selection. Picked Python + Streamlit as the stack — Streamlit because it produces a serviceable web UI from a single Python script and matched the team's familiarity with form-style tools, and Python because it has the document libraries (python-docx) needed for .docx template filling. HTML/CSS layered in only where Streamlit's defaults needed adjustment.
Data integration. Took each .docx template, identified its variable fields, and wired Streamlit form inputs to those fields. Validated that the generated documents were byte-identical in structure to what the team produced by hand.
Build. Implemented the platform in two surfaces:
- Automation page — staff pick a date, pick one of the nine document types, fill the few fields the system can't derive, and download the rendered .docx.
- Monitoring page — shows which staff member is currently on shift, mapped from the existing roster.
Test & training. Live testing on shift with MEWS staff, presentation + handover on 11 September, runbook for the lead.

Results

9 documents automated end-to-end. Every mandatory daily MEWS report can now be generated in seconds instead of typed from scratch.
Built and shipped in nine weeks. Solo, on-site, including requirements, build, training, and handover.
Operator monitoring layer. The MEWS lead now has a live view of who is on duty without crossing divisions to ask.
Adopted on-shift. Staff used the platform during user testing and accepted it before handover. (Long-term usage status post-handover wasn't formally tracked, so I'm marking that as unverified rather than claiming it.)

Lessons learned

Walking through the workflow beats reading the spec. The first week of sitting with MEWS staff surfaced shadow steps that weren't in the original brief — fields filled from a clipboard pinned to the wall, defaults that everyone "knew" but no one wrote down. Skipping that week would have shipped the wrong tool.
Boring stack > clever stack. Streamlit isn't elegant, but it ships a working form-driven web app in days, not weeks. For a solo nine-week internship that's the right trade.
Handoff is the project, not a postscript. A two-month intern building anything operational is a failure mode unless the team can run it after the intern leaves. The runbook took as long to write as some of the document modules.