HackStack

Project Info

VibeRight

Devpost

Inspiration

We hear a lot about AI and vibe coding. People are building startups with it. But there’s a gap. SENIOR ENGINEERS use VIBE CODING BETTER than JUNIORS. Why? Because they’ve accumulated years of reading code. They can spot issues instantly, debug faster, and anticipate edge cases before they explode in production. Seniors are compounding. Juniors are overwhelmed. They stress over debugging vibe coding with no concrete conception of what their code does. We are NOT just talking about debugging. We are talking about sustainability and scalability of YOUR code so you can vibe code better.

What it does

We are here to fill the Vibe GAP. VibeRight allows learners to become owners of that code without forcing them to manually inspect every line of code. It helps them: review with structure, take ownership, test with intention, catch scalability issues early, ship with confidence. The goal isn’t to slow creativity. It’s to make creativity deliverable.

How we built it

VibeRight has two core components: validating and understanding. 1. Validating — A Multi-Agent Expert System Instead of one general critic, we built specialized agents, each obsessed with one dimension of code quality: Functional agent → Your code runs, tests pass, no hidden debt. API contract agent → Robust APIs. Validation, errors, security boundaries. And bulletproof dependency management. Architecture agent → Clean separation of concerns. No unmaintainable spaghetti. Concurrency agent → No race conditions. No intermittent production failures. Performance agent → Scales efficiently. Algorithmic bottlenecks caught early. Security agent → Injection attacks, unsafe execution, auth vulnerabilities : blocked before prod. Documentation agent → Code others can actually understands. Resilience agent → Graceful failures. Recovers, doesn't crash silently. Quality agent → Readable. Maintainable. Low cognitive load. Dependency agent → Supply chain protected. Vulnerabilities, outdated packages, licensing risks are flagged. Each agent produces structured feedback. Then we cluster, merge, and score the outputs into actionable quality metrics. 2. Understanding — Active Ownership Training Validation isn’t enough. So we built a chat-based learning loop powered by two LLMs: An Interviewer that quizzes you on your own code. An Evaluator that checks whether you actually understand it. The questions go deeper over time . You refine your answers. You correct misconceptions. Step by step, you build durable knowledge in an accessible environment. We’re not just reviewing code. We’re training long-term engineering intuition.

Challenges we ran into

Defining “Good Vibe Code”. How do we define those agents and decide what makes vibe code good without impairing creativity? Unlike LeetCode, there’s no fixed input-output pattern. We wanted to preserve creativity, not constrain it. So how do you enforce quality without standardizing thinking? We had to define expert mental models, not rigid rules. Another challenge is SCALE — how can these agents, even when parallelized, handle huge files and interconnected folders/repos effectively?

Accomplishments we're proud of

We turned an abstract frustration into a working system we actually use. We are proud to see the concept come to life. It’s something we can actually use to become better builders. We focused on real pain points shared by all coders. We debated hard. We disagreed productively. We made structured decisions. And we built something non-obvious.

What we learned

We started trying to "validate code", we ended up learning to capture how experts actually think. Our main takeaway: don't build one smart critic, build 10 specialized ones: each focused on a single thing like security or performance. This catches what a general tool misses, and keeps feedback fast enough to not break your flow. We also learned that how you ask is as important as what you build. The right prompts are basically your system design. We mostly learned a lot about ourselves, and how to divide and conquer when we're tackling building abstract concepts. We found our own pattern for tough problems : find someone who does it well, break down how they think, then automate that. We learned to spend compute where it matters.

What's next

VibeRight expands to all users, feedback and comments on each other's performance and code. Challenges on new concepts. Online vibe hackathons where you can be as creative as you want, without intimidation from bulk programs. We can see this having an impact on education. We believe this changes how we learn to code: less theory, more building, faster feedback loops. New platform to assess student skills. A LeetCode for the AI era. VibeRight becomes a platform for technical interviews preparation + mock interviews that test how you think, and use AI. Students and career-changers get immediate, quality feedback that used to take years of senior mentorship. Catch up fast. Ship faster.

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only
  • AnthropicIn code
  • CSSIn code
  • FlaskIn code
  • JavaScriptIn code
  • Next.jsIn code
  • PythonIn code
  • ReactIn code
  • Tailwind CSSIn code
  • TypeScriptIn code
  • OpenAIClaimed

9 of 10 appear in the indexed code. 1 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

  • CodexConfig
  • CursorConfig · Commits

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

357 KB

Source files

106

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.

0 stars