Project Story: Wikikwiki

About the Project

Picture this: you're curious about something, anything really, and you do what we all do...you go to an LLM or Wikipedia. But that same topic you just read about? It's got a completely different story depending on which language you're searching for it in.

I kept running into this all the time, and it was driving me absolutely nuts. I'd stumble across these incredible Japanese wiki pages packed with historical details, or French entries with technical explanations that made everything click, or I'd translate Spanish articles that gave me cultural context that completely reframed my understanding. Meanwhile, my English searches across multiple LLMs or Wikipedia felt like they were barely scratching the surface.

So I had to ask: what if we're missing out on all this incredible knowledge just because we only search in one language?

That's where Wikikwiki comes in. It's my attempt to crack open this multilingual treasure chest and make all that hidden knowledge actually findable.

What Inspired Me

The initial spark came when I was hungry one night, and jonesing for ramen. As I sat down for dinner, I thought I'd pull up the wiki and learn a little more about the history of the dish. To my surprise, the Japanese Wikipedia entry was nearly triple the length of the English version and included a much deeper dive into ramen's culinary history, regional differences, and even political context that were all missing from the English page.

I started to search and the more I did, the more I found this wasn't just about noodles. This goes deeper than that. Japanese politics were barely covered in Spanish. Benefits programs for migrants weren't translated outside of English. And wars... well, let's just say they're written by the winners.

I wanted to create a tool that could extract and expose this deep, multilingual knowledge in a way that was simple, searchable, and accessible.

What I Learned

I learned how fragmented open knowledge truly is across languages. And how sometimes certain controversial topics may actually be glossed over or lack depth in certain languages.

Technically, I deepened my understanding of Wikipedia’s API structure, cross-language article mapping, and content parsing strategies. I also had to consider linguistic nuances that make direct translation unreliable without deeper analysis.

How I Built It

Wikikwiki connects a custom frontend to a backend that leverages Wikipedia’s multilingual APIs. Every time a user enters a topic, the system uses AI to search all available language versions of that article. It then filters for the most content-rich source (which is typically the version with the largest file size or highest character count) and translates it using AI to make that deeper knowledge accessible in English.

The interface is designed to evoke the feeling of exploring a hidden archive or activating a knowledge recovery protocol. Styled like a retro-futuristic control panel, it gives users the sense that they’re uncovering buried insights and restoring forgotten data.

Challenges I Faced

• Cross-language linking: Mapping equivalent articles across different languages wasn’t as clean as expected. Wikipedia interlanguage links sometimes fail or point to loosely related content.
• Depth metrics: Measuring “richness” had to go beyond word count. I had to experiment with metadata signals like internal link density, section counts, and update frequency.
• Translation fidelity: Literal translation often fails to capture tone, specificity, or cultural framing. I considered integrating neural translation layers but decided to keep initial versions raw for transparency.
• Design clarity: Presenting this as a functional, legible tool while still retaining its exploratory theme was a balancing act.

Accomplishments that we're proud of

We built a working multilingual search and translation pipeline that dynamically identifies the richest source material across Wikipedia. For the hackathon we're using mock data since AI API costs were prohibitive. We created a visually distinctive interface that makes the process feel immersive and purposeful. Most importantly, we proved that English articles are often just the surface and that there is immense value in mining beyond them.

What we learned

We learned that multilingual knowledge access is both technically feasible and deeply rewarding. We also realized that the structure of Wikipedia itself creates blind spots for monolingual readers. Working on Wikikwiki helped us better understand the limits of mainstream search and how much more is available just a layer deeper.

What's next for Wikikwiki

Next, we want to expand support for side-by-side comparisons between languages, improve translation fidelity with context-aware models, and allow users to request translations on demand. We're also exploring ways to rank or tag topics by how much more content is available in other languages, helping surface the biggest gaps in English knowledge coverage.

Built With

• bolt
• openai
• tailwind
• wikipedia