Challenge Project

GeoPostal Insight

Challenge G - Post office, postal service and digital service mapping platform


⛶  Fullscreen ↓  Download 📂 Demo

0. Context

"Just for fun":

Elevator Paul Challenge
UPU Innovation Challenge 2024 Challenge

The hackathon before the hackathon - not an official challenge!

1. Quick experiments with OSM + LLM

Rohit Gautam's OSM GPT works well for a very quick, LLM-assisted comparison of post office density in OpenStreetMap. Here we zoom in on the capital of Brazil and nearby rural areas:

Here is a complete OSM data export being worked on in the QGIS desktop app:

Screenshot from 20240530 122624.jpg

2. Wireframe and sketch

We tried to put ourselves into the "shoes" of the data analyst for the Post Office in Brasil, and mocked up a user interface (first on paper - see the sketch in the log), then digitally using the Excalidraw tool:


3. AI prototype

Using the new Partyrock experimental product from AWS, we tried feeding a long description from the challenge presentation and our mockup into an automatic app generator. We first had to load the OSM data into QGIS, and export it as a CSV file with X - Y values. The first result was intriguing, as a thought model, but without more work not usable. We will see if we can refine the prompts to focus on generating a more thoughtful analysis.

Try our AI app here: https://partyrock.aws/u/datalets/vRbPv3TRW/Postal-Intelligence-Insights

Screenshot from 20240530 115931.jpg

Screenshot from 20240530 122032.jpg

4. Check in May 30 15:42

  • Seb: A few infrastructure layers (roads, 4G, population) and post offices. Wrapping up Postal Bank data.
  • Paul: Research into data from the Postal Office (Correios), looking into their published survey. Context. Will share link.
  • Julius: Providing data to the group, knowledge about postal network. What kind of approach in pairing up challenges.
  • Siu: Making a Q&A style app for questions, linked to answers database (group A), rethink the parameters of digitalization. Help the operator or regulator make decisions in regards to optimizing the use and collection of data.
  • Rafael: Searching government connectivity (mobile fiber) data, presenting index of activity and facilities. Trying to fit this with the postal data.
  • Gustavo: Demonstrate the correlation between lack of flexibility in regulatory framework and performance of the Post. Oscar is helping us create a dashboard to select different scenarios, show how this compares to other countries. Chatbot with info about any country.
  • Oleg: Looked into scrapability of Correios database, worked a bit with QGIS. A Data Package with a small extract from the UPU statistics portal added to GitHub. Working in Jupyter Hub locally and in AWS Glue to set up an ML pipeline.

5. Check in May 31 12:10

  • Séb: Assisting with further data preparation.
  • Ihar: Identifier supplementary indicators (opened Post Bank accounts), including the ratio of key accounts, for income map account - trust of population to postal service. Data is old but we are showing the approach.
  • Oleg: Set up dashboards using Panel / Jupyter, visualizing the data using charts. Now adding Folium for better geodata rendering. Would like to integrate a quick LLM demo into the UI prototype as well.
  • Paul: Setting up slide presentation, weaving together the story of the problem we're solving with the dashboard.

6. Final presentation

We came up on stage and briefly described the context, and showed our presentation. Due to the notorious "demonstration effect" also plaguing other teams (note to self: make a screencast next time), we could not show our chatbot in action. It is, however, available live through the demo link. Screenshot:

Screenshot from 20240531 171254.jpg

The project development was on the whole quite productive, we look forward to doing some retrospective and tidying up our recommendations for our esteemed UPU & ITU colleagues, and the wider interest group.

GeoPostal Insight

Challenge G: Post office, postal service and digital service mapping platform.

For details visit our dribdat page.


Our project investigates postal services in Brazil as a use case. A Data Package with a small extract from the UPU statistics portal is available here. We are also sharing archival data from the UPU that was provided to us at the hackathon. Please do not assume that any of this is open access: contact UPU for permission to reuse.


Install Miniconda or Anaconda, then:

conda env create -n post -f requirements.txt

conda activate post

Open the root folder in Jupyter to explore our notebooks, or launch in dashboard mode with:

panel serve geopostal.ipynb

In development, you might find the --autoreload parameter useful.

In production, append --allow-websocket-origin <FQDN> where FQDN is your full domain name.


We use NGINX to host the demo site. There's an example configuration in nginx.conf.

This content is a preview from an external site.

Event finished

Edited (version 82)

31.05.2024 15:17 ~ loleg


Screenshot from 20240531 171254.jpg The demo effect is real! Thanks for listening to our pitch, try the chatbot for yourselves 🤖

31.05.2024 15:17 ~ loleg

Edited (version 80)

31.05.2024 14:05 ~ loleg

  We are ready. Thanks everyone ☺️ Go team!

31.05.2024 13:51 ~ loleg

Edited (version 75)

31.05.2024 13:42 ~ loleg

  🎉 Our solution is live deployed at https://geopostal.schoolofdata.ch/ 🎉

31.05.2024 13:39 ~ loleg

Fixed notebook

Edited (version 71)

31.05.2024 13:35 ~ loleg


Posted a team status update

31.05.2024 13:35 ~ loleg

Added geopandas

Merge branch 'main' of github.com:sg-peytrignet/upu-hackathon-challenge-G

remove first header (@Sebastien Peytrignet)

Merge branch 'main' of github.com:sg-peytrignet/upu-hackathon-challenge-G

add pop map (@Sebastien Peytrignet)

Added aggregated GeoJSON

Impact > Insight

Updated requirements.txt

Updated layout

Added Folium map

31.05.2024 11:07 ~ Ihar


  Say "olá!" to our virtual assistant on HuggingFace (choose Try as Guest)

31.05.2024 10:59 ~ loleg

Joined the team

31.05.2024 10:48 ~ Siu

Histograms improved

Added Brazil geodata

Initial dashboard visual


Joined the team

31.05.2024 07:55 ~ Ihar

Edited (version 65)

31.05.2024 06:51 ~ paulsobon

delete solution deck pdf - powerpoint is latest version

31.05.2024 06:51 ~ paulsobon

Edited (version 64)

31.05.2024 06:50 ~ paulsobon

added map of postal points in Brazil as provided on usb drive

31.05.2024 06:50 ~ paulsobon

Edited (version 63)

30.05.2024 16:52 ~ paulsobon

add map of postal points vis-a-vis population

30.05.2024 16:52 ~ paulsobon

Hurray! I'm done processing the geodata for Brazil on: population, road network, cell towers, postal offices and postal banks at the h3 hexagon (resolution 4) level, the file is hosted here https://drive.google.com/file/d/1EmdeSyncdVcVWNbBY_SH_C2o_pqwD2f3/view?usp=drive_link

30.05.2024 15:28 ~ sgpeytrignet

Anyone using Apache Airflow yet to wire up their data sources?

30.05.2024 15:14 ~ loleg

  Should we go up for some air? Nah, let's get more 📦 data + more ☕ coffee!

30.05.2024 15:04 ~ loleg

The output of a NN trained on density mapping with interpolation as described here

30.05.2024 15:01 ~ loleg

  Open Infrastructure Map by Russ Garrett (shout-out to ⚡ EMF Camp!)

30.05.2024 14:25 ~ loleg

Added Brazil statistics via Data Package (made with Creator, pictured) in the GitHub repository

30.05.2024 14:11 ~ loleg

Edited (version 49)

30.05.2024 14:11 ~ paulsobon

Data Package

Edited (version 48)

30.05.2024 14:07 ~ paulsobon


put Solution Pitch deck into the Pitch presentation slot

30.05.2024 14:07 ~ paulsobon
30.05.2024 14:05 ~ paulsobon

Added 2006 statistics

Added statistical data package



Event started

Contributed 2 weeks ago by sgpeytrignet for UPU Innovation Challenge 2024
All attendees, sponsors, partners, volunteers and staff at our hackathon are required to agree with the UN System Code of Conduct. Organisers will enforce this code throughout the event. We expect cooperation from all participants to ensure a safe environment for everybody.

The contents of this website, unless otherwise stated, are licensed under a Creative Commons Attribution 4.0 International License. The application that powers this site is available under the MIT license.