Project Codexa

Project Codexa — Your offline AI tutor, in your language.

Comment

🧑‍🏫 Project Codexa

🌟 Inspiration

The inspiration behind Project Codexa was the idea of turning learning into an interactive journey instead of just passively consuming text or videos. I wanted to build a personal AI tutor that feels encouraging, fun, and adaptive — combining quizzes, games, and explanations to simulate a classroom experience, but available anytime.

What makes Project Codexa special is that it runs completely offline, powered by an open-source GPT-20B model. This means no internet connection is required, giving learners privacy, accessibility in low-connectivity regions, and independence from cloud services — a step toward making AI education tools more valuable for humanity.

📚 What it does

  • Acts as a personal AI tutor with conversational explanations.
  • Generates quizzes inside styled boxes, where users can answer and get instant feedback.
  • Provides interactive games like:
    • Rearranging scrambled Python code
    • Ordering biological processes (DNA → RNA → Protein)
    • Mixing virtual chemicals with reactions
  • Tracks progress and accuracy, awarding badges (🥉 Bronze → 💎 Diamond).
  • Includes Text-to-Speech (TTS) support so the tutor can “speak” responses.
  • Offers a dashboard for subject-wise performance and achievements.
  • Runs completely offline — no internet needed for tutoring, quizzes, or speech.

🛠️ How we built it

  • Streamlit: For chat interface, sidebar, quizzes, games, and dashboard.
  • Regex parsing: To extract structured quizzes from AI tutor responses.
  • streamlit-sortables: For drag-and-drop functionality in the games.
  • pyttsx3: For offline text-to-speech audio generation.
  • GPT-OSS 20B: An open-source 20-billion parameter model that powers the tutor fully offline.
  • Session state management: To persist chat history, quiz stats, and progress across reruns.
  • Gamification design: Badges, balloons, and progress bars to boost learner motivation.

⚡ Challenges we ran into

  • Managing state across Streamlit reruns (avoiding lost progress).
  • Quiz formatting issues: Making sure answers weren’t pre-selected and results displayed clearly.
  • Integrating drag-and-drop games without breaking logic when items scrambled.
  • Audio generation: Ensuring offline TTS quality with pyttsx3.
  • Running large models locally: Optimizing GPT-20B inference speed without internet dependency.
  • Time pressure: Balancing features and polish within limited development time.

🏅 Accomplishments that we're proud of

  • Successfully built a gamified AI tutor that combines chat, quizzes, and games in one place.
  • Designed a system that runs 100% offline, including AI inference and TTS.
  • Created a beautiful quiz format in styled boxes, with correct/incorrect answers highlighted.
  • Designed a fun learning dashboard that tracks progress and awards badges.
  • Integrated interactive educational games that make abstract topics engaging.
  • Delivered a project that feels like both a tutor and a study companion.

📖 What we learned

  • How to effectively use Streamlit’s session state for persistent experiences.
  • Parsing structured responses from AI with regular expressions.
  • The importance of user engagement through gamification.
  • Running and optimizing a GPT-20B open-source model fully offline.
  • Hands-on experience with offline TTS in Python.
  • Using LaTeX rendering in Streamlit for clean math formulas:

[ E = mc^2 ]

🚀 What's next for Project Codexa

  • 🎤 Add speech-to-text so learners can talk to the tutor naturally.
  • 🌐 Multi-language support for global accessibility.
  • 📱 Build a mobile app version with offline learning modes.
  • 🧑‍🤝‍🧑 Enable multiplayer quiz competitions for study groups.
  • 📊 Expand the dashboard with more analytics like streaks, time spent, and subject mastery.

Built With

  • gpt-oss-20b
  • python
  • pyttsx3
  • regex
  • session-state
  • streamlit
Share this project:

Updates