EchoHire: AI-Powered Mock Interviewer

Architecture
Start Screen with selection as Interviewer
Start Screen with selection as Candidate
Interview Screen as Candidate
Interview Screen as Interviewer
Interview Report as Interviewer
Interview Report as Candidate

Inspiration

Job interviews are high-stakes for both sides. Candidates struggle to articulate their value, and interviewers often struggle to ask the right questions to accurately evaluate talent. With the release of Gemini 3.1 Pro, we saw an opportunity to build a dual-mode platform that uses high-level reasoning to coach both the seeker and the hirer, ensuring professional excellence on both sides of the table.

What it does

EchoHire is an adaptive, multi-mode AI career agent. Powered by Gemini 3.1 Pro and Gemini 2.5 Live, it offers two distinct pathways:

The Interviewee Path (Candidate Prep): The AI acts as a sophisticated interviewer. Using the candidate's CV and a Job Description, it conducts a bidirectional voice interview. It uses self-reasoning to adapt question difficulty in real-time and generates a deep performance report.
The Interviewer Path (Manager Training): In a unique twist, the user plays the role of the interviewer while the AI acts as a candidate. After the session, the AI provides feedback not on the answers, but on the quality of the user’s questions, suggesting revisions to help the user extract more accurate data and better evaluate candidates.

How we built it

Frontend Architecture: Built with React.js and Vite, utilizing TypeScript (.tsx) for robust, type-safe development of the dual-mode interface.
Real-time Voice & VAD: Integrated Voice Activity Detection (VAD) to allow for instant interruption. This creates a fluid, bidirectional conversation where the AI stops talking the moment it detects the user's voice.
AI Engine: A multi-model approach using Gemini 3.1 Pro for complex agentic reasoning and Gemini 2.5 Live for low-latency multimodal audio.
Cloud Infrastructure: Entirely serverless deployment on Firebase and Google Cloud (GCP), using Firebase Functions for backend logic and Cloud Firestore for session persistence.

Challenges we ran into

Mode-Switching Logic: Implementing a "reverse" feedback loop where the AI evaluates the user's questioning technique required complex prompt engineering within the 3.1 Pro engine.
Version Locking: We faced "Version Drift" where environments defaulted to legacy 1.5/2.5 models. We implemented strict overrides to ensure the app remained anchored to the 3.1 Pro latest release for superior reasoning.
Technical Pivot: To solve audio capture and reporting issues found in early AI Studio prototypes, we transitioned the entire backend to Firebase Functions, which provided the stability needed for bidirectional voice.

Accomplishments that we're proud of

The Reverse-Coaching Logic: Successfully building a system that can critique a hiring manager’s questioning style to improve candidate evaluation.
Agentic Flow: Moving away from scripted questions to a 100% reasoning-based dialogue with no hard limits on conversation length.
Firebase Stability: Achieving a live, cloud-deployed state that handles high-token reasoning and multimodal audio seamlessly.

What we learned

Infrastructure is Everything: The pivot to Firebase was the turning point for a stable user experience.
The Power of Reasoning: We learned that 3.1 Pro isn't just for answering questions—it’s powerful enough to evaluate the intent and effectiveness behind how questions are asked.
Collaboration via Cloud: Coordinating between local development and Firebase deployment taught us the importance of environment variable security and model pinning.

What's next for EchoHire

Live Video Analysis: Integrating Gemini 2.5 Live Video to provide feedback on non-verbal cues for both interviewers and candidates.
Industry-Specific Personas: Deep-loading sector-specific logic for technical, legal, and executive leadership roles.
Enterprise HR Integration: Building a dashboard for HR teams to standardize interviewing quality across their organizations.