AI for Good: Humanitarians.AI | Research & Software Development
Students turn questions into working prototypes, from a cat-adoption chatbot to AI-generated proteins, under Professor Nik Bear Brown's guidance at Humanitarians AI.
Most classes end when the semester does. This one is framed around the opposite question: what if graduation is the beginning, and the ideas students build in class are meant to keep working in the real world after they leave? That premise runs through a set of very different student projects, from a chatbot for cats to AI-generated proteins, all built under Professor Nik Bear Brown's guidance.
From a hard question to a working prototype
The pattern across every project is the same: start with a genuine question, then build something that answers it rather than just theorizing about it. "Can a chatbot help special needs cats?" became Catbot, a 24/7 matchmaker that pairs senior cats with forever families, built on natural language processing trained specifically to communicate with empathy. Catbot doesn't just answer questions about adoption, it advocates for the animals it's trying to place.
Rethinking how learning itself works
Two projects turn the research lens on learning itself. The Cognitive Type project studies typefaces not as decoration but as cognitive tools, mapping specific fonts to memory and reading outcomes, treating serif choices as something that can give struggling readers "second chances." Lyrical Literacy takes a parallel approach with sound instead of shape, using AI to craft songs, storybooks that sing, and audiobooks that move with rhythm, putting music and neuroscience directly in service of how children read.
Systems that work behind the scenes
Not every project is meant to be seen directly. Some teams built the infrastructure other AI systems run on: multi-agent systems that don't just answer questions but solve problems, find patterns across large datasets, delegate subtasks, and collaborate with each other, described as "a swarm of minds as one." Survey Mind takes a related approach to a different problem, generating synthetic survey respondents as personas rooted in real psychology and modeled on the Big Five personality traits, aimed at the ways real-world surveys fail to capture how people actually think.
Reaching into biology
The most ambitious project described is PredictaBio, where Humanitarians AI created some of the first AI-generated proteins, built by pulling information from millions of research papers and synthesizing it with large language models, then testing the results at the edge of what current biology can validate. It's a clear step beyond software into experimental science, built with the same "ask a real question, then build" approach as the rest of the program.
Key takeaways
- Every project starts from a specific, real-world question rather than an abstract assignment.
- Catbot uses NLP trained for empathy to match senior cats with adopters, functioning as an advocate rather than a simple chatbot.
- The Cognitive Type and Lyrical Literacy projects apply AI to how people read and learn, using typography and music respectively.
- Multi-agent systems built by students handle delegation, reasoning, and pattern-finding behind the scenes of other tools.
- Survey Mind generates psychologically grounded synthetic personas to address weaknesses in traditional survey research.
- PredictaBio pushed furthest into new territory, generating AI-derived proteins from published research literature.
Who this is for
This overview is for anyone curious about what applied, socially motivated AI research actually looks like when it's built by students rather than described in the abstract. It's a direct look at the range of work coming out of Humanitarians AI's programs, spanning software, education, and biology under one shared philosophy: build the thing, don't just discuss it.
Full transcript(auto-generated, with timestamps)
[0:00]What if graduation wasn't the end, but the beginning? What if we built ideas not just in theory, but into the world? What if students didn't wait to be told what matters. They built what does. AI. AI for good. Where learning meets doing. Where genius multiplies. Under Professor Nick Bear Brown's guidance, students turn questions into prototypes, dreams into frameworks, knowledge into action, they asked, "Can a chatbot help special needs cats?" They built Catbot, a 247 matchmaker, pairing senior cats with forever families built on NLP trained in empathy. Catbot doesn't just chat, it advocates. Others looked at letters, not just how type looks, but how type thinks.
[1:00]The cognitive type project mapped fonts to memory. Serifs to second chances at reading. They didn't just create type faces. They asked what makes words stick. And what if kids learn to read by singing? Lyrical literacy used AI to craft songs. Story books that sing. Audio books that move with rhythm. Music met neuroscience. Melody became a classroom. Some built games in 3D worlds where stories breathe, learning becomes play, players become makers. In the back end, teams built the invisible. They built agents and more agents. Agents that didn't just answer. They solved systems, found hidden patterns, and oceans of data. Delegated, reasoned, collaborated. A swarm of minds as one.
[1:45]And in a world where surveys fail, they reimagined respondents. Survey mind made synthetic personas rooted in real psychology modeled on the big five showing us how minds might think. Then came the impossible. What if we could write new proteins just by reading? Predictabio humanitarians AI created the first AI generated proteins pulled from millions of papers synthesized with LLMs tested at the edge of biology. And there were many more projects in branding, storytelling, equity. Each a seed, each a spark. This is more than class. This is AI for good. One part learning, one part doing, all parts possible. From branding to biology, from surveys to singing, from fonts to forever homes
[2:38]And games that teach to systems that think, we are humanitarians. AI
More videos
2:08Bridging the Pixel Gap in Browser Automation.
2:23How One Narrow Safety Rule Can Make an AI Less Safe Everywhere Else.
2:04Why splitting a chunk from its document makes it retrieve for the wrong question
4:20Three You Can Take Back. One You Can't.
2:21Why a 50-turn agent pays for the same screenshot 35 times unless it caches the pixels
1:53