GIS internship portfolio - Kisumu City Hall: satellite water detection (browser, zero-install), parking revenue accountability, pseudo-3D buildings, census data analysis. All public data.
# Kisumu City Hall GIS — Internship Portfolio
**Author:** Randy · GIS intern, City Hall GIS Department, Kisumu, Kenya
**Period:** Week 1 (Aug 2026)
**Constraint that shaped everything:** guest accounts, no admin rights, slow IT bureaucracy → **every tool is zero-install: open in a browser or run from a USB stick.**
---
## Projects at a glance
| # | Project | Status | What it is |
|---|---------|--------|-----------|
| 01 | **Water Body Lab** | ✅ Working | Satellite water/pond/flood detection in the browser (NDWI/MNDWI/Otsu) — streams real Sentinel-2 scenes, zero install |
| 02 | **Parking Accountability** | 🔶 Concept + scripts | Revenue-leakage audit for city parking: reconciliation of payment logs vs physical capacity (not ANPR) |
| 03 | **Pseudo-3D Buildings** | 🔶 Design | Footprint + floor-count pseudo-3D for the city — drone/lidar/gyroscope approaches killed on evidence |
| 04 | **Zero-Install Playbook** | 📘 Lessons | The pattern that survived: anything needing IT involvement dies; anything in a browser/USB survives |
| 05 | **Census Data Analysis** | ✅ Working | Ward-level demographic investigation ("Katito ward" / "Seme ward") — exact cited figures from KPHC 2019 |
## Flagship demo (60 seconds, no install)
1. Open `01-water-body-lab/water_body_lab.html` in Chrome.
2. Sample scene loads automatically → click **Auto threshold (Otsu)**.
3. Blue = water detected; pond count + areas appear.
4. Real data: click **🌐 Sentinel-2 · Kisumu scene** (streams from NASA's open S3 bucket) or upload a GeoTIFF.
## What this week proved (the short version)
- **Positioning matters as much as the tool:** the flood/pond tool was validated by the GIS team as *internal decision-support* (they need faster signal; they can't publish error-prone public maps). That framing is why it survives.
- **Reconciliation beats detection:** parking accountability is a discrepancy problem (reported vs actual), not a "count every car" problem — a fraction of the cost and no surveillance infrast …