VisionAlign

VisionAlign: Go beyond the resume. We use AI to find candidates who align with your company's core vision and culture, not just keywords.

Comment

My Inspiration: Fixing a System That's Failing Everyone

We've all felt it. As a candidate, you pour your heart into a job application, only to have your resume vanish into a digital black hole, likely filtered out by a keyword scanner. As a recruiter, you're buried under a mountain of those same resumes, desperately trying to find a signal in the noise. The truth is, modern hiring is bottlenecked by an outdated process, and it's failing both companies and the incredible talent they're looking for.

This hackathon's call to "build a better future with AI" was the spark we needed. We weren't inspired to just automate the existing process, but to fundamentally change it. We envisioned a solution that uses AI for what it does best: deeper understanding. Our goal was to build a tool that makes hiring more effective, more equitable, and ultimately, more human by focusing on the one thing a resume can't show you: how a person thinks.

What VisionAlign Does: From Keywords to Character

VisionAlign transforms the hiring process from a game of keyword-matching into a genuine conversation about values and vision. Here’s how it works:

  1. AI-Powered Job Creation: A recruiter doesn't just list skills; they describe their "vision" for an ideal team member. Our system, using Mixtral-8x7B-Instruct via the Together AI API, then crafts a complete job description and a set of insightful, open-ended questions designed to reveal a candidate's problem-solving style and personality.

  2. Authentic Candidate Responses: Instead of just uploading a resume, candidates answer these bespoke questions. This gives them a real opportunity to showcase who they are beyond a list of previous jobs.

  3. Our Two-Stage AI Evaluation Engine: This is our secret sauce. To ensure fairness and authenticity, we built a custom evaluation pipeline:

    • Stage 1 (The Gatekeeper): A fast, efficient Llama-3-8B-Instruct model first scans for AI-generated answers, ensuring the responses are genuine and from the candidate themselves.
    • Stage 2 (The Scorer): Once verified, a powerful Llama-3-70B-Instruct model performs a deep, nuanced analysis, scoring the human-written answers against the recruiter's original vision.

  4. A Data-Driven Shortlist: The recruiter gets a clean, simple dashboard of applicants, ranked by their "Vision Alignment Score." This instantly surfaces the best-fit candidates, transforming what could be a week of manual screening into a 5-minute task.

How We Built It: A Modern, Secure, and Production-Ready Stack

We chose our tech stack to directly address the hackathon's challenges, building an application that is secure, scalable, and production-ready from day one.

  • AI Engine *(Fulfilling the "Build with AI" Challenge): We fully embraced the "future of open-source AI" by using the **Together AI API* as our exclusive AI provider. This gave us the flexibility to use the best open-source models for each specific task—a powerful approach we're incredibly proud of.

  • Backend & Security *(Fulfilling the "Security" Challenge): We built a robust serverless backend using **Supabase, including its PostgreSQL database and Edge Functions. This was a critical architectural decision. All AI interactions and logic are handled on the server, meaning **our API keys are 100% secure and never exposed on the client-side.*

  • Frontend *(Fulfilling the "User Experience" Challenge): We created a polished, intuitive, and fully responsive UI using **React, TypeScript, and Tailwind CSS*, ensuring a seamless experience for both recruiters and candidates.

Challenges We Ran Into (And How We Solved Them)

Our biggest challenge was building an AI system that was not just powerful, but also reliable. The raw output from LLMs can be unpredictable. We overcame this by:

  1. Enforcing Structured Output: We engineered our prompts and used the `response_format: { "type": "json_object" }` feature in our API calls. This forces the LLMs to return clean, predictable JSON, which eliminated parsing errors and made our entire backend far more stable.

  2. Building a Secure AI Gateway: As mentioned, we consciously architected our app to route all AI interactions through Supabase Edge Functions. This wasn't the easiest path, but it was the right one. It ensures our application is secure by design, a principle we were unwilling to compromise on.

Accomplishments We're Proud Of

  • Fulfilling the Hackathon's Vision: We are incredibly proud of building a project that we feel directly answers the main "Build with AI" challenge while also aligning perfectly with key sponsor tracks like Together AI, and the themes of security and production-readiness.

  • Our Multi-Model AI Engine: Our two-stage "Gatekeeper" and "Scorer" system is a sophisticated solution that goes beyond a simple API call. It's a thoughtful approach that ensures both the integrity and the quality of our results.

  • A Complete, Polished Product: We pushed ourselves to deliver a full-featured application, not just a proof-of-concept. From the AI-powered backend to the secure, user-friendly frontend, VisionAlign is a complete and functional platform that's ready for real users today.

What We Learned

This hackathon was a masterclass in building practical AI applications. Our biggest takeaway is the importance of using the "right tool for the right job." Using a small, fast model for simple detection and a large, powerful model for nuanced reasoning was far more effective and efficient than a one-size-fits-all approach. We also solidified our skills in building secure, serverless applications that are ready for real-world deployment.

What's Next for VisionAlign

Our vision for this project extends far beyond the hackathon. We believe this is the future of hiring. Our next steps are:

  1. Launch on Product Hunt: To gain initial user traction and valuable feedback.
  2. Build out our Analytics Suite: Provide companies with data-driven insights on their hiring process, helping them refine their vision over time.
  3. Develop Key ATS Integrations: Create an API to plug VisionAlign's scoring engine into platforms like Greenhouse and Lever, allowing us to scale rapidly within existing enterprise workflows.

We believe VisionAlign can create a world where who you are matters more than the keywords on your resume. Thank you for considering our project.

Built With

  • netlify
  • react
  • supabase
  • togetherai
Share this project:

Updates