DEV Community: Sakharam Thorat The latest articles on DEV Community by Sakharam Thorat (@sakharam_thorat_26b3a6702). https://dev.to/sakharam_thorat_26b3a6702 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3879672%2F245c2331-0d0b-4ad1-9270-3e508d023626.png DEV Community: Sakharam Thorat https://dev.to/sakharam_thorat_26b3a6702 en How We Built Aura: A Real-Time AI Interview Coach with Gemini Live, ADK, and Vertex AI Sakharam Thorat Wed, 15 Apr 2026 05:03:56 +0000 https://dev.to/sakharam_thorat_26b3a6702/how-we-built-aura-a-real-time-ai-interview-coach-with-gemini-live-adk-and-vertex-ai-1g38 https://dev.to/sakharam_thorat_26b3a6702/how-we-built-aura-a-real-time-ai-interview-coach-with-gemini-live-adk-and-vertex-ai-1g38 <p>This article was created for the purposes of entering the Gemini Live Agent Challenge. #GeminiLiveAgentChallenge</p> <h2> πŸŽ™ Try Aura Live </h2> <p>🌐 Live Demo: <a href="https://aura-sde-interview-agent-ivhauk7c7a-uc.a.run.app" rel="noopener noreferrer">https://aura-sde-interview-agent-ivhauk7c7a-uc.a.run.app</a><br> πŸ’» GitHub Repo: <a href="https://github.com/srthorat/aura-sde-interview-agent" rel="noopener noreferrer">https://github.com/srthorat/aura-sde-interview-agent</a></p> <h2> The Biggest Barrier Isn’t Skill. It’s Interview Fear. </h2> <p>Many talented candidates get shortlisted for top tech companies, but never reach their full potential in interviews.</p> <p>Not because they lack technical ability.</p> <p>Because they lack realistic interview practice.</p> <p>Mock interviews are expensive, difficult to schedule, and often inaccessible for students outside elite networks or major cities.</p> <p>Most preparation resources are text-based.</p> <p>People read. Few truly practice.</p> <p>We wanted to change that.</p> <p>So we built <strong>Aura</strong> β€” a real-time AI interview coach powered by Google AI models and Google Cloud.</p> <p><strong>No text box. Just an interview.</strong></p> <h2> ⚑ What is Aura? </h2> <p>Aura is a voice-first interview simulator that lets candidates practice technical interviews naturally through conversation.</p> <p>Candidates can:</p> <ul> <li>Practice coding rounds</li> <li>Practice system design rounds</li> <li>Practice behavioral interviews</li> <li>Receive spoken feedback instantly</li> <li>Get live grading updates</li> <li>Return later and continue where they left off</li> </ul> <p>Aura behaves like a real interviewer β€” not a chatbot.</p> <h2> 🧠 Why Voice-First Matters </h2> <p>Interviewing is a spoken skill.</p> <p>You need to think aloud, structure answers, handle pressure, communicate tradeoffs, and respond in real time.</p> <p>Typing into a textbox cannot simulate that.</p> <p>Aura uses live voice conversation so candidates can build confidence under realistic conditions.</p> <h2> πŸ— System Architecture </h2> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fv37tkujxf06v5i80dinh.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fv37tkujxf06v5i80dinh.png" alt="Aura Architecture" width="800" height="533"></a></p> <h3> Core Stack </h3> <p><strong>Frontend</strong></p> <ul> <li>React + Vite</li> <li>LiveKit JS SDK</li> </ul> <p><strong>Transport</strong></p> <ul> <li>LiveKit WebRTC for low-latency audio streaming</li> </ul> <p><strong>Backend</strong></p> <ul> <li>FastAPI (Python 3.11)</li> <li>Containerized on Cloud Run</li> </ul> <p><strong>AI Layer</strong></p> <ul> <li>Gemini Live (gemini-live-2.5-flash-native-audio)</li> <li>Google ADK agent orchestration + tools</li> </ul> <p><strong>Memory Layer</strong></p> <ul> <li>Vertex AI session persistence</li> </ul> <p><strong>Infra</strong></p> <ul> <li>Cloud Build</li> <li>Artifact Registry</li> <li>Secret Manager</li> </ul> <h2> πŸ”Š Why We Chose Gemini Live </h2> <p>We wanted natural, interruptible conversation.<br> Gemini Live gave us:</p> <ul> <li>Real-time bidirectional voice streaming</li> <li>Fast spoken responses</li> <li>Interruption / barge-in handling</li> <li>Tool calling during conversation</li> <li>Native voice experience without stitching multiple systems together This made Aura feel like a real interviewer.</li> </ul> <h2> πŸ›  How Google ADK Helped </h2> <p>We used Google ADK to structure interview workflows with reusable tools such as:</p> <ul> <li><code>evaluate_candidate_answer</code></li> <li><code>submit_rubric_grade</code></li> <li><code>get_round_scorecard</code></li> <li><code>get_session_summary</code></li> <li> <code>end_conversation</code> This let us combine model intelligence with deterministic scoring logic.</li> </ul> <h2> πŸ’Ύ Persistent Candidate Memory with Vertex AI </h2> <p>Aura supports returning candidates.<br> It can remember:</p> <ul> <li>Previous interview rounds</li> <li>Prior scores</li> <li>Weakness areas</li> <li>Suggested next practice topics</li> <li>Improvement over time This transforms Aura from a one-time mock interviewer into a long-term coaching platform.</li> </ul> <h2> ☁️ Why Cloud Run </h2> <p>We wanted fast iteration and production-ready deployment.<br> Cloud Run gave us:</p> <ul> <li>Containerized deployments</li> <li>Autoscaling</li> <li>Low operational overhead</li> <li>Easy CI/CD integration</li> <li>Reliable public hosting for demos Perfect for hackathon speed and startup scalability.</li> </ul> <h2> πŸš€ What We Learned </h2> <ol> <li>Voice UX matters more than UI polish</li> <li>Latency is everything in real-time AI</li> <li>Persistent memory creates user stickiness</li> <li>Interview confidence is trainable</li> <li>Google AI stack dramatically accelerates development</li> </ol> <h2> 🌍 Bigger Vision </h2> <p>Aura starts with interview coaching.<br> The same architecture can power:</p> <ul> <li>Sales roleplay coaching</li> <li>HR screening simulators</li> <li>Language learning tutors</li> <li>Customer support training</li> <li>Certification prep agents</li> </ul> <h2> Closing </h2> <p>We believe the future of learning and preparation will be conversational, adaptive, and always available.<br> Aura is one step toward that future.</p> <p><strong>Aura</strong><br> <strong>No text box. Just an interview.</strong></p> <ul> <li>#GeminiLiveAgentChallenge #GoogleCloud #Gemini #VertexAI #AI #VoiceAI #Hackathon</li> </ul> ai geminiliveagentchallenge machinelearning gemini