A full-stack AI assistant that helps job seekers with resume feedback, mock interview preparation, and career advice using Retrieval-Augmented Generation (RAG) and Model Context Protocol (MCP).

📺 Watch the Full Demo on YouTube - 15-20 minute walkthrough covering design decisions, architecture, and live demonstration of all features.

This project focuses on Career Coaching and Job Search Assistance, a domain that is particularly well-suited for RAG + tools because:

1. Rich Knowledge Base: Career guidance requires access to extensive, up-to-date information about resume writing, interview techniques, industry trends, and career transitions
2. Practical Tools: Job seekers need actionable tools like resume analysis and mock interview generation that can provide personalized feedback
3. Dynamic Content: Career advice needs to be current and relevant, making RAG essential for retrieving the most recent and applicable guidance
4. Personalization: Different career stages and industries require tailored advice, which tools can provide based on user input
5. Measurable Impact: Career coaching has clear success metrics (job placement, interview success, salary negotiation outcomes)

• Resume Analysis: Get detailed feedback on your resume with specific improvement suggestions using LLM-enhanced analysis
• Mock Interview Preparation: Receive role-specific interview questions and preparation tips
• Career Advice: Access curated career guidance from indexed documents using RAG
• Enhanced Chat Interface: Intelligent chat with automatic tool selection and multiple formatting options
• Streaming Chat: Real-time streaming responses with Server-Sent Events for better user experience
• RAG Pipeline: Semantic search across 20+ career documents for relevant advice
• MCP Server: FastMCP-based tools for seamless AI interactions
• Modern Web UI: Beautiful React frontend with Material UI
• Multi-LLM Support: Gemini (default) with OpenAI fallback
• ReAct Pattern: Transparent reasoning and tool usage

• Docker and Docker Compose installed
• Google API key for Gemini (recommended) or OpenAI API key

cp env.example .env
# Add your Google API key to .env for Gemini (recommended)
# Or add OpenAI API key as fallback

# Production mode
docker-compose up -d

# Development mode (with hot reload)
docker-compose -f docker-compose.yml -f docker-compose.override.yml up -d

• Frontend: http://localhost:3000
• Backend API: http://localhost:8000
• API Documentation: http://localhost:8000/docs
• Health Check: http://localhost:8000/health

docker compose down

1. Install Dependencies:

   pip install -r requirements.txt

2. Index Documents:

   python rag/indexing.py

3. Start MCP Server:

   python mcp_server/server.py

1. Install Dependencies:

   cd frontend
   npm install

2. Start Development Server:

   npm run dev

• Document Indexing: 20+ PDFs/markdown files on resume writing, interview strategies, and career advice
• Vector Storage: FAISS for efficient similarity search
• Chunking Strategy: Semantic chunking with overlap for better context preservation
• Multi-query Reformulation: Enhanced document retrieval through query expansion
• Advanced Technique: Multi-query reformulation with paraphrasing models

• Tool 1: analyze_resume(resume_text) - Comprehensive resume analysis with LLM enhancement
• Tool 2: mock_interview(position) - Role-specific interview questions and preparation tips
• Resource: career_guides://featured - Curated career tips using RAG
• Chat Endpoint: /chat - Main interface with ReAct pattern implementation

• Default: Google Gemini 1.5 Flash
• Fallback: OpenAI GPT-4o-mini
• Prompt Engineering: ReAct pattern with structured reasoning
• Error Handling: Graceful degradation with rule-based fallbacks

• React 18: Modern React with hooks
• Material UI: Professional UI components
• Vite: Fast build tool and dev server
• Responsive Design: Works on all devices

User: "Can you analyze my resume?"
Assistant: 
Thought: User needs resume analysis. I should call the analyze_resume tool.
Action: analyze_resume(resume_text="[user's resume]")
Observation: Tool returns detailed feedback on structure, content, and improvements
Final Answer: Here's your comprehensive resume analysis with specific improvements...

User: "I have an interview for a data scientist position"
Assistant:
Thought: User needs interview preparation for a specific role. I should call the mock_interview tool.
Action: mock_interview(position="data scientist")
Observation: Tool returns role-specific technical and behavioral questions
Final Answer: Here are likely interview questions for a data scientist role...

User: "How do I transition from marketing to data science?"
Assistant:
Thought: User needs career transition advice. I should retrieve relevant information.
Action: retrieve(query="transition from marketing to data science")
Observation: RAG returns career transition and skills development guides
Final Answer: Here's a step-by-step plan to transition from marketing to data science...

User: "Can you help me with my resume?"
Assistant: I'd be happy to analyze your resume! Please paste your resume text below, and I'll provide detailed feedback on structure, content, and improvements.

User: [pastes resume text]
Assistant: 
## Resume Analysis Results

**Analysis Type:** llm_enhanced
**Resume Length:** 1,247 characters

### Feedback:
- Strong use of action verbs: developed, managed, created
- Consider adding more quantifiable achievements
- Good structure with clear sections
- Skills section could be more specific to your target role

---
*Analysis completed using AI-enhanced tools*

The enhanced chat interface automatically:

• Detects user intent and selects appropriate tools
• Supports markdown, plain text, and code formatting
• Maintains conversation context
• Shows which tools were used for transparency

User: "Can you help me prepare for a data scientist interview?"
Assistant: [Streaming response appears in real-time]
🤔 Analyzing your request...
🎯 Detected intent: INTERVIEW_PREP
⚙️ Generating interview questions for data scientist...
📝 I can help you prepare for interviews! What position are you interviewing for? I'll generate relevant questions and preparation tips.
📊 Tools used: ['mock_interview']
🎯 Action: request_interview_position
✅ Streaming completed!

The streaming chat interface provides:

• Real-time responses: See responses as they're generated
• Progress indicators: Visual feedback during processing
• Stream cancellation: Stop responses mid-stream
• Smooth UX: No waiting for complete responses
• Server-Sent Events: Efficient streaming protocol

project/
├── README.md
├── requirements.txt
├── env.example
├── docker-compose.yml
├── Dockerfile.backend
├── data/                 # 20+ Career documents (markdown)
├── rag/
│   ├── indexing.py      # Document indexing and vector storage
│   └── retrieval.py     # RAG retrieval with multi-query reformulation
├── mcp_server/
│   ├── server.py        # FastMCP server with tools and resources
│   └── llm_client.py    # Multi-LLM client (Gemini + OpenAI)
├── prompts/
│   └── rag_prompt.py    # ReAct pattern prompt engineering
├── frontend/            # React frontend
│   ├── src/
│   ├── package.json
│   └── Dockerfile.frontend
├── evaluation/
│   ├── test_queries.json # 10 test queries for evaluation
│   └── results.md       # Comprehensive evaluation results
└── examples/
    └── demo.md          # Example interactions with ReAct pattern

The system has been comprehensively evaluated with:

• Average Relevance Score: 4.4/5
• Tool Accuracy: 93.5%
• RAG Precision: 78%
• Average Response Time: 2.3 seconds

• 10 Realistic Test Queries: Covering resume analysis, interview preparation, career transitions
• Manual Scoring: 1-5 scale for relevance and quality
• Precision Metrics: Precision@3, Precision@5, Recall, F1 Score
• Failure Mode Analysis: 4 identified failure modes with mitigation strategies

• Strong performance in core capabilities
• Good tool selection accuracy
• Effective RAG retrieval for career advice
• Robust fallback mechanisms when LLM unavailable

• Backend: Python, FastAPI, FAISS, SentenceTransformers
• Frontend: React 18, Material UI, Vite
• LLM: Google Gemini 1.5 Flash (default), OpenAI GPT-4o-mini (fallback)
• Vector Database: FAISS
• MCP Framework: FastMCP
• Prompt Engineering: ReAct pattern with structured reasoning
• Containerization: Docker, Docker Compose

1. Add markdown files to data/ directory
2. Rebuild the backend container: docker-compose build backend
3. Restart: docker-compose up -d

1. Make changes in frontend/src/
2. Changes will auto-reload in development mode
3. For production, rebuild: docker-compose build frontend

1. Modify mcp_server/server.py
2. Changes will auto-reload in development mode
3. Check API docs at http://localhost:8000/docs

• Set GOOGLE_API_KEY in .env for Gemini (recommended)
• Set OPENAI_API_KEY as fallback
• Change DEFAULT_LLM in .env to switch providers

• Port conflicts: Ensure ports 3000 and 8000 are available
• API connection: Check that the backend is running and accessible
• Document indexing: Verify that markdown files are in the data/ directory
• Environment variables: Ensure .env file is properly configured
• LLM availability: Check API keys and service status

# View all logs
docker-compose logs

# View specific service logs
docker-compose logs backend
docker-compose logs frontend

curl http://localhost:8000/health

MIT License - see LICENSE file for details.