Clarus

Inspiration

As AI grows stronger and stronger, it is more and more tempting to just copy/paste code and blindly trust ai to build you full stack apps. The trade off with that is you don't actually understand the code nor do you build a habit of read the code explanations the ai gives you. To fix this issue we would create our own IDE called Clarus to be your AI helper. important question: Are we building developers who truly understand systems, or simply users who generate output? We noticed that many learners rely heavily on AI to "Vibe code" without fully understanding what the code is doing. The result is a fast progress, but often shallow learning. Under the education track, we wanted to rethink how AI fits into the learning journey. Instead of replacing engineering thinking, we asked: How can AI guide users to think like real engineers? This question became the foundation of *Clarus *.

What it does

Clarus is a multi-model AI coding assistant designed to combine productivity with real learning. Instead of just generating code, Clarus acts as an intelligent companion that helps users build, improve, and understand their programs. This IDE doesn’t only write code, it teaches the workflow, syntax, reasoning, use cases, and logic behind how the code works. We have currently provide 4 agents: Code buddy - Using your cursor, choose a section of code you want explained and our LLM will break it down for you.
Reviewer - Reviews code and detects for any weaknesses/vulnerabilities. Vibe coder - Generates code for you based on your prompt Rubber Ducky - This agent is to help you debug. You speak to it and the software listens, interprets and walks you through how to debug using elevenlabs AI voice. This is so that you actually learn to debug instead of AI fix it for you.

Vibe coding Agent: Generates structured code from natural language prompts, allowing users to quickly prototype ideas and turn concepts into working solutions. Reviewer and Enhancement Mode: Analyzes generated or existing code to Improve readability, suggest optimizations, refactor inefficient logic. (WE USED MLH BEST USE OF GOOGLE GEMINI) This helps users see how production-level code differs from basic generated output. Selected Text Explanation: Users can highlight specific lines or blocks of code and receive clear, contextual explanations. This transforms passive code usage into active learning. Speech-to-Text: Users can speak their ideas instead of typing them. The system converts speech into text prompts, making coding more accessible and faster for brainstorming . (WE DID MLH BEST USE OF ELEVENLABS) Text-to-Speech: AI explanations can be converted into speech, making the system more accessible and interactive. It supports different leaning styles and improves engagement.

How we built it

We designed using a modular and scalable architecture. The system consists of:

A frontend coding workspace for real-time interaction
Backend APIs to manage different operational modes
A large language model (LLM) API for code generation and analysis
Speech processing APIs for both Speech-to-Text and Text-to-Speech Each user request follows a structured workflow: User input -> Mode Selection -> AI Processing -> Structured Output

Challenges we ran into

trying to configure the gemini api key and continually getting a 403 error of api key leak, when pushing to git. -Trying to connect all API keys for our IDE agents.

Accomplishments that we're proud of

Building a functional multi-mode AI assistant within hackathon constraints.
Successfully integrating review, enhancement, explanation, and voice features.
Designing a system focused on learning rather than pure automation.
Creating a seamless workflow between frontend, backend, and AI services.

What we learned

To build an IDE and understand the components behind one.
Make sure the idea is bullet proof before your start even thinking about building it. ## What's next for Clarus ?
Many small features such as fine tuning LLM models to give better responses.
Getting investors to continue building this project so we can help more people become developers.