AI-Interview-Pro

• 

  Interview Agent System Workflow

• 

  Multi Modal System Workflow

Inspiration

I am a rising junior, and knowing how competitive the engineering field is, I have to prepare myself for getting internships and projects. In order to do that, I need an amazing resume and be able to breeze through the interviews.

What it does

AI-Interview-Pro is an end-to-end career platform. It transforms messy text into ATS-optimized resumes, auto-generates LinkedIn profiles, matches users to real jobs using semantic search, and conducts real-time multi-modal (video/audio) mock interviews with an AI agents, allowing for it to give actionable feedback.

How we built it

Frontend: React 18 and Vite with a custom UI. We used native browser APIs for real-time video/audio capture. Backend: Python and FastAPI for asynchronous data handling. AI Stack: Google Gemini API (core reasoning & parsing), Sentence Transformers (local semantic job matching), ElevenLabs (TTS), and OpenAI Whisper (STT).

Challenges we ran into

API Rate Limits: Handling quotas (like HTTP 429s) forced us to build a robust fallback system to native browser speech APIs when cloud services failed. Structured Outputs: Engineering prompts so the LLM would consistently return perfectly formatted JSON for the resume parser. Real-time Latency: Managing the pipeline of recording audio, transcribing it, generating AI responses, and streaming TTS back to the user without breaking conversational flow.

Accomplishments that we're proud of

True Multimodal Agents: Moving beyond text by integrating live video (face tracking) and audio (speech pacing) so the AI agent can see and hear users in real time. Multi-Agent Orchestration: Successfully chaining specialized agents together, from a Resume Parser structuring raw data to a Job Matcher using semantic search, and finally the Interview Agent. Graceful Resilience: Building smart fallbacks so the agents gracefully downgrade to browser APIs if premium multimodal tools fail.

What we learned

How to efficiently orchestrate multi-step LLM chains and stream data between third-party AI models and a React frontend. Implementing lightweight, local semantic search using Sentence Transformers to give our agent context without an expensive vector database. Mastering browser-based WebRTC and media streaming APIs to build multimodal agents.

What's next for AI-Interview-Pro

Agentic Handoffs: Creating a multi-agent system where a "Resume Agent" hands off context directly to a "Technical Interview Agent" for a specialized coding sandbox experience. Live Job APIs: Allowing the agent to autonomously scrape and recommend live listings from platforms like Indeed or LinkedIn. Progress Analytics: A dashboard to track user confidence scores and filler-word usage over time.

Built With

• ai
• elevenlabs
• fastapi
• gemini
• javascript
• multimodal
• openai
• python
• react
• tailwindcss
• vercel