NeoSearch

NeoSearch — The AI-powered compass for French-speaking tech learners. Find the right resources, get personalized guidance, and accelerate your growth — all in one place.

Comment

Inspiration

NeoSearch was born from a simple but persistent frustration: Despite the vastness of the internet, French-speaking developers waste hours searching for relevant, high-quality, and accessible content in their own language.

We noticed:

Scattered and fragmented tech resources

A language barrier that limits access to English-only content

A lack of personalized tools for learners of different skill levels

No intelligent guidance to support structured growth

What it does

NeoSearch is the first AI-powered search engine tailored for the French-speaking tech community. It helps users find the right resource at the right time, whether it's a tutorial, hackathon, project, or training course — and adapts to their skill level and learning goals.

Key features:

Fast, intelligent multi-source search (hackathons, tutorials, GitHub, certifications)

Video mentor powered by Tavus, with realistic AI avatar interaction

Gamification system with dynamic badge tracking

Age-based learning paths (student, adult, senior)

Personalized and contextual AI assistant

Offline-ready state saving and session continuity

How we built it

Frontend: React + Vite + TypeScript for speed and scalability

Database: Initially MongoDB, but migrated to Supabase/PostgreSQL for better security, full-text search, and built-in auth

Backend/API: Supabase Edge Functions, custom parallel search logic with timeout fallback

AI Mentor: Integrated Tavus video avatar (with fallback simulation mode)

Search engine: Parallel real-time fetches from curated sources, intent detection system, and caching layer

Gamification engine: Activity-based progression system using user behavior and search frequency

Challenges we ran into

Tavus integration: Complex API auth, fallback design, video compression for performance

Search latency: Sequential fetching caused timeouts — had to build parallel logic with fail-safe mechanisms

State persistence: Ensuring search history isn’t lost when navigating externally

Intent detection: Building a robust NLP-like system to understand user queries (over 15+ intent types)

Localization: Adapting all logic to French culture and language while keeping the platform in English

Architecture shift: Moving from MongoDB to Supabase mid-hackathon for better performance and structure

Accomplishments that we're proud of

Successfully integrated a realistic AI mentor using Tavus

Reduced search response time by over 70% through parallelization and caching

Designed an engaging gamification system that boosts user activity

Built a platform that feels human, responsive, and intelligent

Achieved a 23% user signup rate and 4.2 average searches per session in demo testing

Created a tool that genuinely serves an underrepresented tech audience

What we learned

Relational databases with built-in auth (like Supabase) are more efficient for structured learning apps

Video avatars require resilient design with offline fallback

TypeScript and modular architecture are vital for maintainability under time pressure

User engagement increases drastically with gamified elements

Localization goes beyond translation — it means adapting content, tone, and journey to the user’s reality

What's next for NeoSearch

Integrating Machine Learning to improve search result relevance

Developing a React Native mobile app

Launching a public API for developers and partners

Expanding to other language markets like Spanish and Italian

Adding voice search enhancements and real-time learning recommendations

Built With

Share this project:

Updates