Cloo | Devpost

Inspiration

Roughly 1 in 7 residents of metro Atlanta speaks a language other than English at home — over 660,000 Spanish speakers alone, plus large Vietnamese, Korean, West African, and South Asian language communities with deep, decades-long histories in this specific city. For many of these residents, breaking into tech isn't just about learning to code — it's about parsing job postings, error messages, and interview questions written in dense, unexplained English jargon. We built Cloo so that language is never the reason someone doesn't get into tech.

What it does

Cloo takes confusing tech content — a job posting, an error message, an interview question, or a general term — and returns a plain-English explanation, a translation into the user's language, a practical next-step tip, and related terms worth knowing. A Mock Interview mode generates role-specific interview questions and gives bilingual feedback on typed answers. A Study Deck lets users save and review explained terms as flashcards, stored entirely client-side. Explanations support follow-up questions and an adjustable reading level, and text-to-speech playback supports pronunciation across nine languages, including three Ghanaian and West African languages rarely supported by mainstream language tools. A "Facts about ATL" page grounds the whole product in the real immigrant and refugee communities of metro Atlanta.

How we built it

Frontend: React (Vite) with Tailwind CSS v4, framer-motion, lucide-react, and react-hot-toast. Backend: Node.js with Express, serving both the API and the built frontend as a single Render Web Service. AI: Google Gemini API (gemini-3.5-flash-lite) for explanation, translation, and interview logic, with Google Cloud Text-to-Speech and Speech-to-Text extending voice features beyond browser-native support. No database — Study Deck data persists client-side via localStorage, keeping the architecture simple and fast to ship within the hackathon window.

Challenges we ran into

Gemini model selection took real trial and error: gemini-2.5-flash returned 404s as no longer available to new API keys, and the full gemini-3.5-flash model was usable but slow due to default "thinking" behavior; thinkingLevel: "MINIMAL" combined with the flash-lite model fixed this. We also hit a genuine language-coverage gap: neither browser-native speech APIs nor Google Cloud support Twi at all, despite it being central to Cloo's identity. We researched and identified Khaya AI, a Ghana-based research API built specifically for underserved West African languages, as the real fix — full integration didn't make it into this build, so rather than hide the gap, the app degrades honestly, hiding voice features only where no real backend supports them. We also made deliberate scope cuts under time pressure, including a community forum and a mentorship marketplace, after concluding they had low real value without an existing user base.

Accomplishments that we're proud of

Shipping a genuinely full-featured product in one day: explanation, translation, mock interviews, voice input/output, a study tool, and a locally-grounded content page, not just a single AI-wrapper feature. We're proud that instead of quietly hiding our one real language gap (Twi voice support), we researched it honestly and found a legitimate specialist solution rather than pretending a mainstream provider covered everything. We're also proud of the visual identity — a deliberate Ghana-inspired red/black/gold theme and a custom "Cloo" wordmark, not a default template look.

What we learned

How uneven "global" language support actually is across major cloud AI providers — Twi, spoken by millions, simply isn't in Google's speech models, while a small, purpose-built research API (Khaya AI) covers it well. We also learned a lot about scoping under real time pressure: cutting features with weak ROI (a forum, a mentorship marketplace) early freed up time for features that actually strengthened the core product, like mock interviews and the reading-level toggle.

What's next for Cloo

Closing the Twi voice gap with Khaya AI's dedicated Ghanaian-language API. Context-anchored community threads attached to specific explained terms, rather than a general forum. Expanded language coverage that treats regional providers as first-class options instead of assuming one cloud vendor covers every language equally well.