Today, I’m sharing my recent project: CareerGPT.
CareerGPT is built using LangGraph and LangChain. The core idea behind this project is to design a multi-agent system where different specialized agents handle different career-related tasks—coordinated by a central supervisor agent.
How It Works
The system consists of 5 specialized agents, all controlled by a supervisor agent.
The supervisor agent:
- Receives the user query
- Analyzes the intent
- Routes the query to the most relevant agent
Each agent is designed with its own prompts, tools, and responsibilities.
The 5 Specialized Agents
- Career Advisor
Provides structured career guidance and suggests clear career paths.
- Resume Analyzer
Users upload their resumes, and the agent analyzes them to provide detailed feedback and improvements.
- Resume Q&A
Allows users to ask questions directly about their resume (e.g., strengths, gaps, improvements).
- Learning Path Agent
Generates custom learning roadmaps based on the user’s target domain (e.g., Web Development, AI, etc.).
- Job Search Agent
Helps users find job opportunities based on specific roles, domains, or companies.
How Agents Differ Internally
Each agent is not just a prompt—it has its own logic and tools.
Job Search Agent
Connected to a web search tool. When a query is received, it searches online (e.g., job listings) and returns relevant results such as available roles in a specific region or domain.
Learning Path Agent
Uses RAG (Retrieval-Augmented Generation) to generate personalized learning roadmaps based on stored domain-specific knowledge.
This separation ensures that each agent is optimized for its specific task instead of trying to do everything in one place.
Tech Stack
- Language: Python
- AI Frameworks: LangGraph, LangChain, RAG
- LLM: Groq (fast and free for experimentation)
- Backend: FastAPI, PostgreSQL, Async Programming, Authentication
- Deployment: Render, Vercel, CI/CD
Feedback
I’d genuinely appreciate your feedback.
If you see gaps, design flaws, or areas for improvement, feel free to point them out—I’m actively looking to refine and improve this system.