CareerForge

Inspiration The inspiration for CareerForge came from my own frustrating experience of hunting for internships. I was spending hours on Jobright and scrolling through endless YouTube career advice videos but everything felt too generic. Even the best videos offer advice meant for a massive audience rather than focusing on my specific background as a data science student. I realized that the most effective career guidance should be a Socratic experience that challenges you based on your unique resume and goals. I wanted to build something that provides personalized mentorship at scale helping people get advice that is actually catered to them rather than a broad group.

How I Built It I built CareerForge using a multi agent architecture I call the Career Quartet. I used the Fetch.ai uAgents framework to orchestrate four distinct agents including Scout for market research Tailor for resume optimization Socrates for interview coaching and Maverick for salary negotiation.

The brain of the project is Google Gemini 1.5 Flash which handles the complex reasoning for each agent. I built the backend with FastAPI to handle asynchronous communication between the agents and used React for the frontend to create a clean interactive user interface. To make the Socrates agent feel like a real coach I implemented streaming responses so the interview practice felt like a live conversation.

Challenges I Faced One of the biggest challenges was managing the state between the frontend and the multiple AI agents running in the background. Since I had four different agents potentially talking to each other ensuring the data flowed correctly from the Scout agent to the Maverick agent without high latency was difficult.

I also ran into some technical hurdles with API rate limits and zero quota errors when I first started testing. I had to spend a lot of time debugging the message passing logic in the uAgents framework to make sure the Socrates coach did not lose context mid interview.

What I Learned This project taught me a lot about agentic workflows and how to bridge the gap between a data science layer and a functional full stack app. I learned how to optimize prompts for specific personas like making Maverick aggressive in negotiations while keeping Socrates helpful and educational. Most importantly I realized that while GenAI is powerful the real value comes from how you structure the interaction to solve a specific personal pain point like the internship search.

Built With

• fastapigoogle-gemini-1.5-flash
• fetch.ai
• javascript
• node.js
• python
• react