🧠 Nexus AI — Technical Interview Platform

AI-powered interview prep with real-time voice, code execution, resume parsing, and grounded web search.

---

✨ Features

Feature	Detail
🎙️ Live Voice Interview	Real-time bidirectional audio via Gemini Live API — interruptions, transcripts, and PCM audio pipeline
📄 PDF Resume Upload	PDF sent as native inline data to Gemini — no client-side parsing, full comprehension
🖼️ Image Attachments	Attach screenshots, diagrams, or whiteboard photos for visual context
💻 Code Workspace	Monaco-style editor with live code execution across 8 languages via Piston API
🔍 AI Code Review	Structured per-language rubric scoring (JS, TS, Python, Java, Go, React, Algorithms)
🌐 Grounded Web Search	Gemini + Google Search grounding for real-time fact retrieval per question
🧠 Session Memory	sessionStorage-backed context: facts from each exchange are stored and recalled
🤖 Dual Persona	Toggle between strict Interviewer and encouraging Tutor mode
⚡ DeepSeek Fallback	If Gemini errors, automatically falls back to DeepSeek-V3 via HuggingFace Router
🗺️ Maps Grounding	Geographic context queries grounded via Google Maps

---

🏗️ Architecture

Browser (React + Vite)
│
├── App.tsx                  — Main UI, file upload, voice/text orchestration
├── services/
│   ├── geminiService.ts     — Gemini API (text, live audio, image gen, code review)
│   ├── mcpClient.ts         — MCP layer: real sessionStorage memory, Gemini search, GitHub API
│   └── compilerService.ts   — Code execution via Piston API (emkc.org)
├── utils/
│   ├── ragEngine.ts         — RAG orchestration: retrieves and records session context
│   └── audioUtils.ts        — PCM encode/decode for Live API audio pipeline
└── components/
    ├── CodeEditor.tsx        — Code editor + run/review panel
    ├── TerminalMessage.tsx   — Chat message renderer (markdown, images, grounding links)
    └── AudioVisualizer.tsx   — Real-time waveform animation during voice mode

---

🚀 Run Locally

Prerequisites: Node.js 18+

# 1. Clone and install
git clone https://github.com/Pavithran200412/Meet-AI
cd Meet-AI
npm install

# 2. Configure environment
cp .env.example .env
# Fill in your API key (see below)

# 3. Start
npm run dev

---

🔑 Environment Variables

Create a .env file in the project root:

# Required — Get from https://aistudio.google.com/apikey
API_KEY=your_gemini_api_key

# Optional — Enables DeepSeek fallback + image generation fallback via HuggingFace
HF_TOKEN=your_huggingface_token

# Optional — GitHub REST API (for candidate repo fetching; public API works without this)
GITHUB_TOKEN=your_github_token

Warning

GITHUB_TOKEN and other secrets are server-side only and must NOT be added to vite.config.ts's define block — Vite inlines define() values as literals into the client bundle. Only API_KEY and HF_TOKEN are intentionally client-visible (this app has no backend).

---

🛠️ Tech Stack

Layer	Technology
Frontend	React 19, TypeScript, Vite 5, Tailwind CSS
AI — Chat & Reasoning	Gemini 2.5 Flash / Gemini 3.1 Pro Preview
AI — Voice	Gemini 2.5 Flash Native Audio (Live API)
AI — Image Generation	Gemini 2.5 Flash Image, Imagen 3
AI — Fallback	DeepSeek-V3 via HuggingFace Router
Code Execution	Piston API (Python, JS, TS, Java, Go, Rust, C++, SQL)
Context / MCP	sessionStorage memory + Gemini grounded search
Deployment	Netlify

---

🎮 Usage Guide

Text Interview Mode

1. Type a message or upload your PDF resume via the 📎 button
2. The AI reads your resume and begins a structured technical interview
3. Questions adapt based on your experience, covering algorithms, system design, and your stack

Voice Interview Mode

1. Tap the 🎙️ mic button to start a live session
2. Speak naturally — the AI interrupts, asks follow-ups, and transcribes in real-time
3. Tap END to stop the session

Code Workspace

1. Navigate to the Workspace tab
2. Write your solution in the editor (supports 8 languages)
3. Hit Run to execute against the Piston sandbox
4. Hit Review to get AI-powered rubric scoring

---

📁 Supported File Types

Type	Handled As
.pdf	Native Gemini inline data — full document comprehension
Images (.jpg, .png, .webp, etc.)	Visual context passed alongside the prompt
.txt, .js, .ts, .py	Raw text injected as file context

---

🔧 MCP Context Layer

The app uses a lightweight Model Context Protocol implementation running entirely in-browser:

• Memory — Each AI exchange stores a summary fact in sessionStorage, recalled on future turns
• Web Search — Queries use Gemini's Google Search grounding for live web results
• Rubrics — Per-language structured scoring rubrics (auto-selected by topic)
• GitHub — Public repo fetching via GitHub REST API