DEV Community: Maribel L. Santos

CatColonyWatch: Local Open AI for Community Cat Field Observation

Maribel L. Santos — Thu, 08 Oct 2026 21:12:56 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

I built CatColonyWatch, an open-source field-observation tool for volunteers and organisations that monitor community-cat colonies.

Its core idea is simple:

The AI is not watching the cats. The human is.

Community-cat monitoring happens outside: at feeding points, streets, gardens, courtyards, shelters, and other shared spaces.

But volunteers often need to do two things at once: pay attention to the animals and keep useful records.

CatColonyWatch is designed to make the screen the shortest part of that experience.

The workflow is:

Go to the colony or feeding point.
Observe the cats and their surroundings.
Record short factual notes.
Let a local open-weight AI model structure those observations.
Review the result before saving it.

The goal is not to automate animal observation.

The goal is to let people spend more time actually observing animals and less time organising notes.

CatColonyWatch currently supports:

reusable records for known cats
unidentified cats observed during a visit
observation start and end times
automatically calculated duration
total cats observed, even when not every cat can be identified
individual notes for known cats
food and water observations
feeding-point conditions
biodiversity observations
visit history
JSON and CSV export
local SQLite storage
AI-assisted structuring with Gemma

The AI has deliberately narrow boundaries.

It does not diagnose animals, assess medical urgency, recommend treatments, or infer emotions or intentions.

Its job is much simpler:

help humans organise what they actually observed.

Demo

Streamlit demo:

https://catcolonywatch.streamlit.app/

The Streamlit version demonstrates the field-observation workflow and interface.

The complete AI workflow is intentionally designed to run locally through Ollama + Gemma 3 on the user's computer.

That means the public interface and the local AI architecture serve slightly different purposes:

Streamlit makes the workflow easy to explore.
Local Ollama + Gemma handles the AI processing while keeping field observations off third-party AI servers.

The full local setup is documented in the repository.

Code

The complete project is open source:

Maribele / CatColonyWatch

🐾 CatColonyWatch Observe first. Record second. Let open AI structure the notes CatColonyWatch is an open-source field-observation tool for volunteers and organisations that monitor community-cat colonies The AI is not watching the cats. The human is.

The app is designed to keep screen time short: volunteers observe cats and their environment first, record factual notes second, and then use a local open-weight AI model to structure those observations into a clearer record 🌿 How it works

Go to the colony or feeding point.
Observe for a few minutes.
Record the cats actually seen.
Add factual notes about individual cats, the environment, and other animals.
Let local open-weight AI structure the observations.
Review the AI output before saving or exporting it. AI supports the fieldwork. It does not replace observation. 🐈 Main features CatColonyWatch includes:

a reusable database for known cats
field visits with date, time, observer, weather and duration
individual…

View on GitHub

Repository:

https://github.com/Maribele/CatColonyWatch

The repository includes:

app.py
requirements.txt
Streamlit interface
SQLite storage
local Gemma/Ollama integration
setup instructions
responsible-use limitations
MIT License

Local databases and secrets are excluded from Git so real colony data is not accidentally published.

How I Built It

CatColonyWatch is built with:

Python
Streamlit
SQLite
Ollama
Gemma 3 4B
Pandas
Requests

The architecture is intentionally small:


text
Human field observation
        ↓
Streamlit interface
        ↓
Recorded visit
        ↓
Local Ollama server
        ↓
Gemma 3
        ↓
Structured JSON
        ↓
Human review
        ↓
SQLite / JSON / CSV

VetBrief: Turning Worried Pet-Owner Stories into Clearer Veterinary Intake Notes

Maribel L. Santos — Sun, 04 Oct 2026 21:24:29 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built VetBrief, a lightweight veterinary intake and appointment request tool designed to help pet owners communicate more clearly with veterinary clinics.

I built it with a friend and veterinary-clinic context in mind.

When people are worried about an animal, they often explain what is happening in a long or unstructured way:

“She has been hiding since yesterday, she normally sleeps on the sofa, she did not eat breakfast, and she walked away when I tried to touch her.”

All of that information may be useful, but veterinary staff need to identify the key facts quickly.

VetBrief helps organise that information before the appointment.

The app collects:

basic patient information
main concern
onset and progression
appetite
water intake
vomiting
stool
urination
respiratory signs
activity and movement
additional observations
known medical conditions
medication and supplements

It then creates a structured veterinary brief.

For additional free-text observations, VetBrief uses an open-weight language model to turn the owner's description into a short, factual veterinary intake note.

The owner can then prepare an appointment request with a preferred date and time and attach the veterinary brief.

The goal is not to create an AI veterinarian.

The goal is much narrower:

help people organise what they have observed before speaking with a veterinary professional.

VetBrief does not diagnose, assess urgency, recommend treatment, or replace veterinary judgement.

Demo

Try VetBrief here:

https://vetbrief-maribele.streamlit.app/

The demo lets you:

complete a veterinary intake form
create a structured brief
use open-weight AI to structure free-text owner observations
prepare an appointment request

Code

GitHub repository:

https://github.com/Maribele/vetbrief

The repository includes:

the Streamlit application
setup instructions
dependency requirements
documentation
responsible-AI limitations
instructions for configuring the Hugging Face token securely

How I Built It

VetBrief is built with:

Python
Streamlit
Requests
Hugging Face Inference Providers
an open-weight language model
Git / GitHub

The structured intake form handles information that is easier and safer to collect directly through predefined fields.

The AI component is intentionally limited to one task:

transforming additional free-text owner observations into a concise, factual intake note.

For example:

Owner description

She has been hiding under the bed since yesterday. She normally sleeps on the sofa. She did not eat breakfast this morning and walked away when I tried to touch her.

AI-structured note

Hiding under the bed since yesterday (normally sleeps on the sofa)
Did not eat breakfast this morning
Walked away when approached for touch

The prompt explicitly instructs the model not to:

diagnose
assess urgency
recommend treatment
infer diseases
infer emotions or causes
invent information

That separation is important to the design of the project.

AI structures the information. Veterinary professionals interpret it.

Why Does Open Innovation Matter?

Open innovation matters for VetBrief because this kind of application may eventually handle sensitive information about animals, households, routines, and medical history.

Using an open-weight model makes it possible to explore future versions that could run locally inside a veterinary clinic.

That could give clinics greater control over:

privacy
deployment
model choice
data handling
cost
customisation
evaluation

A closed API can be convenient, but it gives less control over how the model is deployed and how the system may evolve.

For a veterinary context, I think the ability to inspect, compare, adapt, and potentially run models locally is especially valuable.

Open innovation also makes it easier to think critically about the role of AI.

In VetBrief, the model is not treated as an invisible decision-maker.

It is a replaceable component performing a narrow task.

That makes it easier to evaluate whether it is actually useful and whether another open model might perform better.

Prize Categories

Open-source AI / open-weight model