accessAI

Access to AI

Comment

As a Full Stack Developer, I'm thrilled to present accessAI—a project born from the conviction that the best way to learn is by doing.

As Steve Jobs famously said, "Everybody in this country should learn to program a computer, because it teaches you how to think."

Here is the story of accessAI.

Inspiration

The inspiration for accessAI comes from a simple yet powerful observation: passive learning doesn't stick. You can watch tutorials all day, but until you break code, fix bugs, and grapple with logic, you aren't truly programming. We wanted to democratize this "hands-on" experience.

We built this project because everyone deserves access to AI to learn programming, and the only way to truly master it is by dirtying your hands. We wanted to move beyond static tutorials and create a dynamic, AI-powered mentor that pushes users to write code, solve unique problems, and get immediate, constructive feedback.

What it does

accessAI is an intelligent, gamified platform that serves as a personalized programming coach. It doesn't just give you answers; it challenges you to think.

  • Adaptive Learning Paths: Users can choose from curated paths in JavaScript, Python, Java, C++, and Solidity , progressing from basics to advanced concepts like "Asynchronous JavaScript" or "Smart Contracts".

  • Dynamic DSA Problem Generator: Instead of a static list of questions, accessAI generates 100% original Data Structures and Algorithms problems on the fly. You can select your preferred style (LeetCode, Codeforces, AtCoder) and difficulty.

  • AI-Powered Code Review: Users can paste their code to receive a comprehensive review covering correctness, time/space complexity analysis (e.g., ), and edge cases.

  • Intelligent Model Switching: To ensure reliability, the platform integrates multiple AI models (xiaomi, qwen, google/gemini-2.0-flash-exp, Mistral, etc.) and intelligently switches between them if quotas are reached.

  • Gamification: Users earn XP, level up, and compete on a leaderboard, making the "grind" of learning fun and addictive.

How we built it

We architected accessAI as a robust Full Stack application, leveraging the power of modern web technologies and Generative AI.

  • Frontend: We built a responsive, dark-mode enabled interface using React and Tailwind CSS. We utilized react-markdown and react-syntax-highlighter to render beautiful, readable code blocks and lesson content.

  • Backend: The core logic resides in a Node.js and Express server . We implemented secure authentication using JWT and Bcrypt to protect user data.

  • AI Orchestration: This is the secret sauce. We integrated Google's Generative AI (Gemini) and OpenRouter.ai to access a suite of powerful models. We wrote specialized system prompts to act as "Expert DSA Instructors" and "Competitive Programming Setters" to ensure high-quality output.

  • Database: We used MongoDB with Mongoose to store user profiles, progress, achievements, and custom "DSA Progress" maps.

Challenges we ran into

  • API Rate Limits & Quotas: One of our biggest hurdles was hitting the request limits for free-tier AI models. We frequently encountered 429 Too Many Requests errors. To solve this, we implemented a robust API Error Handler that detects quota limits, extracts retry times, and suggests alternative models to the user.

  • Designing Correct Architecture/Flow: Orchestrating state between the learning roadmap, the chat interface, and the problem generator was complex. We had to ensure that the AI "remembered" the context of the current lesson while treating each problem generation as a unique event.

  • Prompt Engineering: Getting the AI to generate valid unique problems without hallucinating impossible constraints required several iterations of refining our system prompts.

Accomplishments that we're proud of

  • Multi-Model Resilience: We successfully built a system that doesn't crash when one AI provider goes down or hits a limit; it simply offers the user a switch.

  • The "Dirty Your Hands" Philosophy: We successfully implemented features that force interaction—you can't just read; you have to generate a problem, write code, and get it reviewed.

  • Math & Code Rendering: seamlessly integrating LaTeX support for mathematical complexity notation (like ) alongside syntax-highlighted code blocks.

What we learned

  • Quota Management is Key: We learned that building AI apps isn't just about the prompts; it's about managing the "fuel" (tokens and requests) efficiently.
  • User Experience in AI: We discovered that raw AI outputs need significant parsing and styling (like our Markdown renderers) to be truly useful for education.
  • The Value of Fallbacks: Relying on a single model is a single point of failure. Diversifying our model sources made our app production-ready.

What's next for accessAI

Currently, the generated data, content, and lessons are ephemeral; every time a user loads a topic, they get a fresh generation. While cool, this burns API credits. We are working on a caching layer to save generated lessons and problems to the database. This will allow users to revisit their favorite lessons without an API call, saving resources and speeding up the experience!

Built With

Share this project:

Updates