Coded in Culture: The Diaspora Technologists Building AI That Finally Gets It

Ask a major AI chatbot to generate an image of a "traditional family gathering" and see what comes back. Ask it to explain why someone might say "I'm dead" in response to a funny tweet. Ask it to recommend music for a Lagos house party or explain the difference between being "pressed" and being "salty." The results, if you're Black, are often somewhere between mildly annoying and genuinely alienating.

This isn't a bug report. It's a structural critique. The datasets that train most mainstream AI systems skew heavily toward Western, white, and English-language-dominant sources. The teams building them are famously non-diverse. And the result is technology that treats Black culture — in all its regional, diasporic, multilingual complexity — as an edge case at best, an error to be corrected at worst.

A growing community of diaspora technologists has decided to stop filing the complaint and start building the alternative.

The Problem With "Neutral" AI

There's a persistent myth in tech that AI is objective — that it processes data without bias because it processes data without feelings. But AI learns from humans, which means it inherits human blind spots at scale. When the training data doesn't include you, the model doesn't know you. And when the model doesn't know you, it makes you up — usually badly.

"I've had AI tools describe Afrobeats as 'African tribal music,'" says Kofi Mensah, a Ghanaian American software engineer based in Atlanta who has spent the last two years building a music recommendation algorithm specifically trained on African and diaspora music catalogues. "These are systems that can write you a sonnet but can't tell Burna Boy from Fela Kuti. That's not neutrality. That's a specific kind of ignorance."

Mensah's project, currently in beta, was born from frustration with streaming platform algorithms that kept serving him the same five Afrobeats tracks regardless of what he actually wanted to hear. He started building his own recommendation engine — one trained on a much broader and more nuanced dataset that includes highlife, Afropop, Afro-fusion, amapiano, and the genre-blending work of diaspora artists who don't fit cleanly into any of those categories.

"The mainstream platforms treat 'African music' like it's one thing," he says. "My model knows the difference between what you want at 2pm on a Tuesday and what you want at midnight on a Saturday. It knows context. It knows culture."

Training Data Is Political

The deeper you go into AI development, the clearer it becomes that training data is never neutral — it's always a set of choices about whose knowledge counts, whose language is legible, whose culture is worth modeling.

For diaspora communities, this has real consequences. Natural language processing tools that don't understand African American Vernacular English misfire on customer service bots, hiring algorithms, and content moderation systems — with documented real-world harms. Image generation tools that default to Eurocentric aesthetics erase representation at scale. Mental health chatbots that don't understand cultural context around family, community, and spirituality can actively mislead the people using them.

Tade Akinola, a Nigerian British developer now based in Oakland, has been working on a conversational AI tool designed specifically for first and second generation African diaspora users navigating the American healthcare system. The tool needs to understand not just English but the specific ways diaspora communities talk about health, body, and medicine — which often includes vocabulary and frameworks that don't map neatly onto clinical language.

"When someone says 'my body is not balanced,' that means something very specific in Yoruba wellness frameworks," Akinola explains. "A generic AI assistant either doesn't understand that or pathologizes it. We're building something that understands it the way a community health worker from that background would."

Beyond the Chatbot

Not all of this work is about fixing existing tools. Some diaspora technologists are building entirely new categories of AI application — ones designed from the ground up around Black diaspora needs, aesthetics, and ways of knowing.

In New York, a collective of Black and Afro-Latina developers called Sankofa Tech Lab is building a cultural recommendation engine trained on a dataset they've spent two years curating — pulling from diaspora media, community publications, music archives, and oral history projects. The goal isn't to build another Netflix or Spotify. It's to build a discovery tool that understands the specific texture of diaspora cultural production and can help users find work that actually speaks to them.

"We're not trying to compete with the big platforms," says collective co-founder Imani Clarke. "We're trying to fill the gap they deliberately leave. There's a whole universe of diaspora content — films, podcasts, music, writing — that the algorithm will never surface for you because it doesn't know it exists or doesn't know you'd want it. We're building the tool that knows."

Other projects are working on AI tools for diaspora business owners — systems trained to understand the specific dynamics of immigrant entrepreneurship, rotating credit associations, and community-based commerce. Some are building language models that work across multiple African languages and their diaspora variants. Others are focused on creative tools — image generators that default to Black aesthetics, music production AI that understands African rhythmic structures.

Who Owns the Model?

There's a thornier question underneath all of this, and the diaspora tech community is actively wrestling with it: who should own AI systems built on Black cultural knowledge?

The concern isn't hypothetical. There's a long history of Black cultural production being extracted, commercialized, and profited from by entities that had no relationship to the communities that created it. The risk that diaspora-trained AI tools get acquired by the same tech giants they were built to counter is real.

Akinola is deliberately building his healthcare tool as a community-owned cooperative. Mensah is exploring open-source licensing structures that would prevent commercial acquisition without community consent. The Sankofa Tech Lab collective has written community ownership into their founding documents.

"We're not building this so Google can buy it in three years," Clarke says flatly. "We're building it for us. The ownership structure has to reflect that from day one, not as an afterthought."

The Agora Goes Digital

There's something fitting about the diaspora — a community that has always had to build its own institutions when mainstream ones failed it — now doing the same thing in the digital infrastructure layer. The agora, the gathering place, the commons: these are not just metaphors. They're requirements.

The AI tools being built by diaspora technologists aren't just technical projects. They're arguments about whose culture deserves to be understood, whose knowledge deserves to be modeled, and who gets to build the systems that increasingly shape how we find information, make decisions, and understand ourselves.

"Every model is a worldview," Mensah says. "We're just finally building worldviews that include us."