HiveMind

Landing Page of HiveMind
List of General Questions
Metrics of User
Company Specific Mock HireVues
Interview Insights

Inspiration

As college students recruiting for competitive internships, we kept running into the same problem - dreadful HireVue interviews with no way to prepare. They're often a bigger filter than in-person interviews, yet somehow the least transparent. Unlike traditional interviews, HireVue questions are extremely time-pressured (with a timer), AI-scored, and nearly impossible to find resources for. We built HiveMind to solve that.

What it does

HiveMind lets anybody prepare for HireVue interviews using real questions sourced from candidates who have already interviewed at top firms like Goldman Sachs, Deloitte, and Accenture. Users practice in a fully simulated HireVue environment with a live camera, timer, and company-specific prompts. During each simulated interview, HiveMind analyzes the user's confidence, facial expression signals, speaking pace, and answer quality, comparing their performance to patterns learned from strong sample interviews.

From there, our dashboard tracks metrics over time, helping users identify weaknesses, monitor improvement, and build confidence through repeated, specialized practice.

How we built it

*Web Scraping and Validation. * Because HireVue questions are proprietary, naive LLM prompting often produced inaccurate or hallucinated questions. To address this, we built a web scraping pipeline that aggregated data from dozens of sources per company based on custom queries, and then used Gemini to extract previously reported HireVue questions from candidate experiences.

Each question, along with its associated roles and companies was iteratively added to a knowledge graph. We tracked how often each relationship appeared across independent sources, using edge frequency as a proxy for credibility. This allowed us to cross-validate questions at scale, filtering out low-confidence entries and retaining only those supported by consistent data.

Frontend. Built with Next.js and React for the user interface, styled using TailwindCSS. The app features dynamic dynamic dashboards, interactive interview practice, real-time analytics powered by live audio and video processing. We integrated lucide-react for clean UI components, Hume AI for emotion and facial expression analysis, Deepgram for live speech transcription, and Gemini API for evaluation.

Challenges we ran into

*Invalid Question Filtering. * Meta-level questions (e.g., "Where can you work" or general advice about HireVue preparation) occasionally appeared in the dataset despite not being actual interview prompts. To address this, we applied cosine similarity filtering against a set of interview questions, ensuring the final dataset remained focused on highly realistic, high-signal interview content.

Accomplishments that we're proud of

Credibility. Rather than relying on LLM-generated questions, which are often inaccurate, we built a model centered on validation and source credibility, ensuring every question is grounded in real data and trustworthy for users.

What we learned

Browser Media APIs. We gained hands-on experience working with browser media APIs, such as MediaRecorder, to capture and process live audio and video streams directly in the browser.

Backend. We learned how to integrate Supabase into our application, handling authentication, database operations, and storage to support user insights.

What's next for HiveMind

Right now, we've been testing with a limited set of companies, roles, and validated interview questions, and the results have been strong. The next step is to expand the breadth of our question bank, increasing coverage across more roles, companies, and industries to support a broader range of users.

We believe in HiveMind so much that we're planning to reach out to the UVA Career Center to use HiveMind as a real-world testbed for students preparing for HireVue interviews. That would allow us to gather quality feedback, validate our approach at scale, and improve the platform with real user data.

HiveMind is building a trusted layer for interview questions. There's a lot of room to grow here, and we're just getting started.