CareerRoguelike- Gamified Job Market Simulator

• 

  Profession Selection - Choose from the list or type anything you want

• 

  Example Task Widget

• 

  Task Completing Success Widget

• 

  Fill in the blanks task widget

• 

  Job Creating Page

• 

  Meeting View

• 

  Graduation Page

• 

  Work Dashboard - Users can see assigned tasks and meetings..

• 

  CV page, When users change their jobs or accomplish tasks the CV is updated

• 

  The sessions are saved to the firebase without login.. But if user wants, can start from the very beginning with any kind of profession

• 

  Interview page with different questions and answers fields

• 

  Job Details Page

• 

  Job Listing page

• 

  Interview Result page with scores questions and answers and the job accept decline buttons

Inspiration

I wanted to create a realistic career simulation that demonstrates advanced multi-agent AI collaboration. The idea was to show how specialized AI agents can work together bidirectionally—where tasks trigger meetings and meetings generate tasks—creating a dynamic workplace experience.

What I Learned

• Multi-agent orchestration: Centralized coordination beats autonomous agents for production systems
• Bidirectional workflows: Agents triggering each other creates emergent, realistic behavior
• Cloud Run scalability: Serverless is perfect for AI workloads with generous timeouts
• Prompt engineering: 40% of development time went into refining AI prompts for consistency
• Pre-validation: Filtering invalid inputs before AI calls saved 30% on API costs

How I Built It

1. Architecture: Designed a Workflow Orchestrator to coordinate 7 specialized AI agents
2. Agents: Built Job, Interview, Task, Grader, CV, Meeting Generator, and Meeting Evaluator agents using Gemini 2.5 Flash
3. Bidirectional System: Implemented intelligent triggers—completing 2-4 tasks generates meetings, meetings generate 0-3 follow-up tasks
4. Frontend: Created React UI with job search, interviews, work dashboard, and virtual meetings
5. Backend: Built FastAPI gateway with direct Gemini API calls for reliability
6. Deployment: Containerized both services and deployed to Cloud Run with auto-scaling
7. State Management: Used Firestore for persistent player state, jobs, tasks, and meetings

Challenges Faced

• Grading consistency: Implemented pre-validation + AI grading to prevent gaming the system
• Task quality: Enhanced prompts to ensure self-contained tasks without external dependencies
• Meeting flow: Built conversation management to detect repetition and determine topic completion
• Dashboard balance: Created intelligent monitoring to maintain 3-5 tasks and 1-2 meetings

Built With

• adk
• docker
• fastapi
• firebase
• firestore
• gemini
• gemini-2.5-flash
• google-artifact-registry
• google-generative-ai-sdk
• googlecloudbuild
• googlecloudrun
• javascript
• nginx
• pydantic
• python3.9
• react
• react-query
• tailwindcss
• typescript
• uvicorn
• vertexai
• vite