NEXUS

• 

  Set your startup idea and budget, then launch the simulation

• 

  All agents collaborate live across research, debate, and execution

• 

  Project summary view

• 

  Agentic card overview

Inspiration

Startups are not built on ideas alone. They succeed or fail because of trade-offs, limited budgets, and conflicting priorities. Product pushes for features, Tech warns about complexity, Marketing drives growth, Finance protects the runway, and Legal flags risk

We wanted to capture that real-world tension with AI. NEXUS simulates an executive team making decisions under constraints, showing what happens when agents argue, allocate capital, and live with the consequences

What it does

NEXUS is a real-time multi-agent startup simulator. You provide a startup idea and a budget. Five AI agents take structured roles and move through clear phases: Research → Proposal → Debate → CEO Decision → Execution. Every approved decision mutates the budget, changes strategy, and has visible financial impact

How we built it

• Built a FastAPI backend to orchestrate phases and lifecycle transitions

• Created DebateManager for multi-round arguments and escalations

• Developed a RAG pipeline with ChromaDB to ground reasoning using curated startup knowledge

• Implemented React + TypeScript frontend to stream events and visualise agent activity, proposals, and budget

• Integrated Stripe sandbox to handle approved spend

• Added ElevenLabs narration for optional live audio feedback

Challenges we ran into

• Coordinating multiple agents with conflicting priorities

• Building a reliable RAG pipeline with ingestion, embedding, and indexing

• Ensuring real-time synchronisation between backend state and frontend visualisation

Accomplishments that we're proud of

• Full multi-agent startup lifecycle instead of a simple AI chat

• Implemented structured debates with voting and escalation

• Built budget tracking and execution tied to decisions

• Developed grounded knowledge retrieval for explainable AI behaviour

What we learned

• Multi-agent systems require strict orchestration

• Grounded retrieval improves reasoning reliability

• Real-time systems demand careful state management

What's next for NEXUS

• Add survival probability scoring and scenario simulations

• Expand the knowledge base

• Add more specialised agents

• Enable session comparison and analytics

Built With

• chromadb
• elevenlabs
• gemini
• postgresql
• python
• react
• ruby
• ruby-on-rails
• stripe
• tailwindcss
• typescript
• vite