Lucent

Inspiration

We've all been there: downloading yet another language learning app, full of excitement on day one, only to watch it collect dust by week two. The problem isn't motivation it's that traditional language learning feels like work. Flashcards, grammar drills, repetitive exercises, progress bars that guilt-trip you into maintaining streaks.

Meanwhile, we spend hours effortlessly consuming content we love podcasts, YouTube, social media, Netflix. Nobody forces us. We just... do it.

That contrast kept nagging at us: What if language learning felt like scrolling through content you actually enjoy?

Then we realized the fundamental flaw in existing apps: they force you to learn their content. Lesson 1, Day 3, Unit 7. Everyone gets the same material whether you care about ordering coffee or debating quantum physics.

What if you decided what to learn about every single day? What if the AI generated an entire immersive experience around YOUR interests, at YOUR level, in YOUR style on demand?

That's when Lucent was born.

What it does

Lucent is a daily language immersion platform where you create your own content. No predefined lessons. No curricula. No "everyone starts with colors and numbers."

Here's how it works:

1. You pick a topic Anything. Today's news, your hobby, a random curiosity. "The psychology behind coffee addiction." "Why my favorite band broke up." "Space tourism ethics." Whatever sparks your interest.

2. You set the vibe Casual like chatting with a friend? Professional? Playful? You control the tone and style.

3. Lucent generates a complete immersion session Using Gemini 3.0 Pro, the app creates:

Listen: A podcast-style audio piece about your topic
Read: An article exploring the theme from a different angle
Speak: An AI conversation about the topic
Write: A prompt that makes you express YOUR thoughts

4. You consume and practice No memorization. No grammar drills. Listen like it's a podcast. Read like it's an article. Talk like it's a voice message. Write like it's a journal entry.

5. Lucent learns from you The AI tracks what you struggle with and naturally weaves practice opportunities into future sessions. You don't even notice you're being taught.

The result? Language learning that feels like your daily content habit not homework.

How we built it

AI Architecture

Gemini 3.0 Pro serves as our primary content engine. We chose Gemini specifically for its exceptional multilingual capabilities it understands linguistic nuances across dozens of languages far better than alternatives. Content generated for learners feels native, not machine-translated.

Gemini 2.5 Flash handles real-time lightweight tasks: instant word definitions, topic suggestions, and feedback generation keeping the experience snappy.

Content Generation Pipeline

We engineered a batched system where a single Gemini 3.0 Pro call produces all module content (Listen, Read, Write prompts) that are thematically interconnected. The Listen piece introduces concepts, the Read piece explores a complementary angle, and the Write prompt synthesizes both creating a cohesive learning journey, not disconnected exercises.

Adaptive Memory System

We built a "memory injection" system that tracks linguistic patterns users struggle with. When generating future sessions, these focus areas are subtly woven into prompts enabling spaced repetition without explicit drilling. The AI learns your mistakes and corrects them naturally, without ever telling you that you made one.

Tech Stack

Next.js 15 with App Router and React Server Components
TypeScript with strict mode
Supabase (PostgreSQL) for persistent user data and session history
Clerk for authentication and subscription management
Tailwind CSS + shadcn/ui for a calm, non-gamified interface

Challenges we ran into

Achieving thematic coherence across modules

We wanted Listen, Read, and Write to feel like one interconnected experience not three random exercises duct-taped together. Achieving this in a single API call while maintaining quality across all outputs was surprisingly difficult. We solved it by giving Gemini explicit instructions about thematic threading and complementary angles.

The tutor paradox

We wanted feedback that helps users improve without making them feel judged. Traditional tutors say "that's wrong." We needed the AI to say "here's how to level up" without using words like "mistake" or "error." Crafting prompts that produce consistently encouraging, coach-like feedback took many iterations.

Avoiding the gamification trap

Every instinct and every competitor screams "add streaks! add points! add leaderboards!" We deliberately resisted. Building an interface that motivates through intrinsic engagement, not guilt and dopamine hacks, remains an ongoing design challenge.

Accomplishments that we're proud of

No predefined content ever. Every piece of content in Lucent is generated fresh for each user, each session. There's no content database. No "Lesson 1." This is radical in the language learning space.

The memory system works invisibly. Users don't realize they're doing spaced repetition. They just keep encountering opportunities to use that tricky grammar pattern. That's exactly the point.

A complete four-skill experience. Most apps focus on one or two skills. We built genuine Listen, Speak, Read, and Write modules that all connect covering the full spectrum of language acquisition.

What we learned

AI changes the economics of personalization. Previously, personalized content meant human tutors ($50+/hour) or pre-made "paths" that still felt generic. Gemini 3.0 Pro lets us generate truly unique content for every user at marginal cost. This fundamentally changes what's possible.

Prompt engineering is an art. Getting consistent, high-quality, level-appropriate content across multiple languages required prompts 10x longer and more nuanced than expected.