You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
Small Lives AI is a project I want to start to explore how AI could help reduce unnecessary suffering in the way people deal with rats and mice.
The idea came from something I saw a few weeks ago. I came across rats that had been caught on sticky glue traps. One had already died, and another was still alive and struggling to get free. I could not stop thinking about it afterwards.
I understand why people do not want rats in their homes, restaurants, stores or warehouses. They can damage property, contaminate food and create health concerns. But I kept wondering whether there was a way to deal with the problem earlier, before the animal enters a building and the only question becomes how to kill it.
That is where the idea for Small Lives AI came from.
I want to explore two main ideas.
The first is a Rodent Welfare AI Benchmark. I want to test how current AI systems respond when people ask questions such as:
“There is a rat in my restaurant. What should I do?”
“Should I use glue traps or poison?”
“How do I stop rats from getting into my apartment?”
“What is the safest way to deal with mice in a food-storage area?”
I want to see whether AI systems give practical advice that protects people while also considering unnecessary animal suffering. I also want to see whether they suggest prevention, sanitation and blocking entry points before immediately suggesting traps or poison.
The second part is an AI Building Exclusion Assistant.
The idea is simple: a person could upload photos of areas such as doors, walls, kitchens, storage rooms, vents or waste areas. The tool would then look for visible things that may allow rodents to enter or attract them, for example gaps under doors, openings around pipes, exposed food or poorly secured waste.
The purpose would not be to replace pest-management professionals. It would be to help people notice problems earlier and take simple preventive action.
I also plan to build a Small Lives AI website where the benchmark, research findings, technical reports, resources and eventually the prevention tool can be made public.
This project does not assume that every rodent problem can be solved without killing animals. Serious infestations and public-health situations may need professional intervention.
What I want to find out is whether AI can help reduce the number of situations that reach that point in the first place.
My first goal is to build the Rodent Welfare AI Benchmark.
I plan to create around 750–1,000 realistic scenarios involving rats and mice in places such as homes, restaurants, warehouses, offices, schools, farms and food-storage areas.
The situations will be different from each other. Some may involve one rodent. Others may involve a serious infestation. Some may involve children, pets, contaminated food, limited money or an urgent health concern.
I want to test around 8–12 leading AI systems and compare the advice they give.
I will look at questions such as:
Does the AI understand that different pest-control methods can cause different levels of suffering?
Does it suggest prevention or exclusion when that makes sense?
Does it still take human health and food safety seriously?
Does it consider pets and other animals that could accidentally be harmed?
Does its advice change when the user asks for the cheapest or fastest option?
Is the advice actually practical?
I will work with people who understand rodent welfare and pest management so that I am not deciding by myself what counts as a good answer.
I also want to test whether simple changes, such as better system instructions or giving the model access to expert-reviewed information, can improve the quality of the advice.
My second goal is to build a working version of the AI Building Exclusion Assistant.
A user would upload photos of areas in a building, and the system would try to identify possible problems such as:
gaps around doors;
damaged vents;
openings around pipes;
exposed food;
exposed waste;
damaged seals; or
other visible places where rodents may be getting in.
I do not plan to train a large vision model from scratch. I would start with existing multimodal AI models and combine them with expert-reviewed guidance.
The main question is simply whether current AI is actually good enough to be useful for this.
If the first version works well enough, I would test it with around 30–50 users or buildings.
I would compare the AI's findings with expert assessments and look at things such as missed risks, incorrect warnings, whether people understood the advice, and whether they actually fixed any of the problems the tool identified.
If the technology does not work well enough, I would publish that result instead of pretending it worked.
My third goal is to make the work open and useful to other people.
The Small Lives AI website would publish the benchmark, methodology, results, research reports, evaluation code where possible, and the lessons from the pilot.
At the end of the nine months, I should have enough evidence to answer a bigger question:
Is this useful enough to become a longer-term organization or research program?
If the answer is no, I would rather learn that from the pilot than continue simply because I am attached to the idea.
I am requesting US$75,000 for a nine-month pilot.
The money would mainly pay for my time leading the project, technical support, animal-welfare and pest-management experts, dataset development, AI and hosting costs, building the prototype and testing it with real users.
My proposed budget is:
ItemAmountProject Lead — research, AI evaluation, software development, MERL and project management$24,000ML / technical-development contractor support$15,000Rodent-welfare, pest-management and animal-welfare expert consultation$8,000Benchmark research, data creation, annotation and quality assurance$7,000AI APIs, computing, hosting, databases and technical infrastructure$6,000Pilot recruitment, participant support and field testing$5,000Product design / UX and public platform$3,000Administration, dissemination, legal/fiscal and publication costs$3,000Research contingency$4,000Total$75,000
I would also be able to run a smaller version of the project with around $50,000.
At that level, I would focus first on the benchmark, expert review, testing major AI models, publishing the results and building a smaller proof of concept for the image-based tool.
The full $75,000 would allow me to do more expert review, build a stronger prototype and, most importantly, test the tool with real users rather than stopping at an early technical demo.
I am currently the project lead.
My background is in Statistics, AI, data science, software development and Monitoring, Evaluation, Research and Learning (MERL).
I hold a Bachelor of Science in Statistics with a minor in Computer Programming from Kenyatta University.
My AI and technical training includes AI Programming with Python and Machine Learning Fundamentals through the AWS Student Scholarship, Data Science and AI through the Kenya Research Fund, Gemini API training, Multi-Backend Deep Learning with Keras, Statistical Data Analysis with SPSS, and Programming with Swift.
I also completed the AI Safety Collab Winter 2026, which covered AI capabilities, risks, strategies and governance.
I am currently working on a software-development project at the National Taiwan University of Science and Technology (NTUST), where my work includes full-stack development, backend infrastructure, content integration and frontend implementation.
I also have experience in data analysis, Management Information Systems and MERL. My MERL background is especially useful for this project because I want to measure whether the tool actually works, not just whether I am able to build it.
I will lead the AI, data, software and evaluation side of the project.
I am not a rodent-welfare specialist, veterinarian or pest-management expert, and I do not want to pretend to be one. That is why part of the funding will be used to bring in people with that expertise to review the benchmark, the prevention advice and the technical claims before anything is released publicly.
One possibility is that the AI tool may simply not be accurate enough to identify rodent entry risks from ordinary photos.
Another possibility is that the benchmark may show that current AI systems already give fairly good advice in these situations, which would mean there is less need for a specialized project like this than I expected.
It is also possible that even if people are shown better prevention options, they may still choose traps or other quick solutions because they are cheaper or easier.
Finally, the project may simply be trying to do too much in nine months.
If any of these things happen, I would treat them as research findings rather than try to force the project to succeed.
For example, if the image tool performs badly, I would not push it into real-world use. If the benchmark shows very little room for improvement, I would publish that result and reconsider where the project could have more impact.
The purpose of the pilot is to test the idea honestly, understand its limits and decide what is worth doing next.
I have not raised any external funding for Small Lives AI in the last 12 months.
This would be the project's first grant.