MentorCrew: Multi-Agent Personalized Mentorship System Inspiration The struggle that junior professionals and students face in obtaining personalized mentorship and constructive feedback throughout their learning journeys. Human mentors are often unavailable 24/7 or prohibitively expensive, creating a gap in continuous professional development. What it does MentorCrew is a collaborative agent system that simulates a team of specialized mentors:
Coach Agent: Sets goals and creates personalized learning roadmaps Technical Reviewer Agent: Reviews code/projects and provides detailed feedback Career Advisor Agent: Suggests career next steps and opportunities Learning Curator Agent: Searches and organizes relevant learning resources Progress Tracker Agent: Monitors evolution and adjusts strategies
How we built it
CrewAI as the main framework for agent orchestration LangChain for memory and context management Vector Database (Pinecone/Chroma) to store knowledge and history LLM APIs (GPT-4, Claude) as the agents' "brain" Streamlit/Gradio for interactive web interface RAG pipeline for domain-specific knowledge retrieval
Challenges we ran into
Maintaining consistency between different agent personalities Balancing agent autonomy vs user control Managing API costs during long mentoring sessions Implementing efficient long-term memory across sessions Preventing agent conflicts when giving contradictory advice
Accomplishments that we're proud of
Creation of distinct and complementary personas for each agent Feedback system that evolves based on user progress Smooth integration between different agents without conflicts Intuitive interface that doesn't expose system complexity Real-time collaboration between agents for complex problems
What we learned
The importance of well-defining each agent's roles and boundaries How to use structured prompts to maintain consistency Caching techniques to optimize costs without losing quality The value of specialized agents vs a generalist agent How to handle agent disagreements constructively
What's next for MentorCrew
Add domain-specialized agents (DevOps, ML, Frontend, Backend, etc.) Implement collaborative learning between users Create marketplace of customizable "agent templates" Integration with real platforms (GitHub, LinkedIn, Coursera, LeetCode) Gamification system with achievements and milestones Voice interaction for more natural mentoring sessions Export learning paths to popular learning management systems
Built With
- crewai
Log in or sign up for Devpost to join the conversation.