🚀 Multi-Agent Startup Intelligence Platform (Novaventure AI)

LIVE DEMO: https://multi-agent-startup-intelligent-sys.vercel.app

An end-to-end AI-Powered Multi-Agent Startup Intelligence & Validation System built with CrewAI, FastAPI, MongoDB Atlas, and React (Vite + Tailwind CSS).

This platform orchestrates a sequential pipeline of four specialized AI agents powered by Llama 3.3 70B (via Groq) and real-time Google Search (via SerperDev) to research market trends, discover viable startup opportunities based on domain and budget, project financial metrics, and validate overall business viability—streaming live execution logs via WebSockets to an interactive analytics dashboard.

📂 Organized Project Architecture

Multiagent_startup_intelligent/
│
├── backend/                        # ⚙️ FastAPI Backend, Database & Parsing Engine
│   ├── api.py                      # Main FastAPI server, CrewAI endpoints & WebSocket log broadcaster
│   ├── db.py                       # Async MongoDB Atlas connection (Motor) & collection accessors
│   ├── report_routes.py            # Report CRUD, ownership checks & aggregated dashboard statistics
│   └── chart_parser.py             # Hybrid JSON/Regex parser extracting structured chart metrics
│
├── security/                       # 🔐 Authentication, Authorization & Security Layer
│   ├── auth_routes.py              # JWT creation/validation, bcrypt hashing, Google OAuth2 & RBAC
│   └── .env.example                # Security & environment variables template
│
├── src/startup_advisor/            # 🤖 CrewAI Multi-Agent Core Engine
│   ├── __init__.py
│   ├── crew.py                     # Agent, Task & Crew definitions (Groq Llama-3.3-70B + SerperDevTool)
│   ├── main.py                     # CLI runner, training, replay & trigger entry points
│   ├── config/
│   │   ├── agents.yaml             # Role, Goal & Backstory definitions for all 4 AI Agents
│   │   └── tasks.yaml              # Task prompts & strict JSON + Markdown output schemas
│   └── tools/
│       └── custom_tool.py          # Custom CrewAI tool templates
│
├── frontend/                       # 💻 Primary Full-Stack React 19 Dashboard (Tailwind v4 + Recharts)
│   ├── src/
│   │   ├── components/             # Interactive charts (ReportCharts.jsx) & Dashboard/Protected layouts
│   │   ├── contexts/               # AuthContext (JWT/Google OAuth) & ThemeContext (Dark/Light mode)
│   │   ├── pages/                  # Dashboard, NewAnalysis, AgentWorkflow, LiveConsole, Report, RunHistory, ProfileSettings, Login, Signup
│   │   └── services/api.js         # Axios client with JWT interceptors
│   ├── package.json
│   └── vite.config.js
│
├── multiagent/                     # 🖥️ Secondary / Modular React UI Workspace
│   ├── src/                        # Domain-specific views (trend, business, financial, validation, profile)
│   └── package.json
│
├── outputs/                        # 📄 Generated AI Agent Markdown Reports
│   ├── report.md                   # Output of Agent 1: Market Trend Analysis Report
│   ├── business_report.md          # Output of Agent 2: Business Discovery & SWOT Report
│   ├── financial_report.md         # Output of Agent 3: Financial & ROI Projections Report
│   └── validation_report.md        # Output of Agent 4: Final Startup Viability & Validation Report
│
├── tests/                          # 🧪 Pipeline & Integration Tests
│   ├── test_agent.py               # End-to-end sequential 4-agent pipeline test script
│   └── test_serper.py              # SerperDev Google Search API connectivity test
│
├── knowledge/                      # 🧠 CrewAI Knowledge Base
│   └── user_preference.txt         # Domain & user context preferences
│
├── uploads/                        # 🖼️ Static User Uploads (Profile pictures served at /api/uploads)
├── .env                            # Local environment secrets (Git-ignored)
├── pyproject.toml                  # Python project metadata & CrewAI scripts configuration
└── README.md                       # Complete project documentation

🤖 The 4-Agent Sequential Intelligence Pipeline

The core intelligence engine (src/startup_advisor/crew.py) runs four specialized agents sequentially, passing context from one stage to the next and appending structured JSON blocks for visual rendering:

| Stage | Agent Name | Role & Responsibility | Tools Used | Output File | Key Extracted Metrics | | :--- | :--- | :--- | :--- | :--- | :--- | | 1 | trend_analyst | Market Trend Analyst: Scans global market trends, emerging technologies, high-demand jobs, and growth sectors. | SerperDevTool (Live Web Search) | outputs/report.md | cagr, top_region, market_size, keywords, historical_growth (5-yr) | | 2 | business_discovery_agent | Business Discovery Specialist: Evaluates the user's target field (industry) and budget to recommend tailored startup models, competitors, and SWOT analysis. | LLM Reasoning (llama-3.3-70b-versatile) | outputs/business_report.md | opportunity_score, competition_level, strengths, weaknesses, opportunities, threats | | 3 | financial_agent | Financial & Growth Advisor: Consumes the Business Discovery report to estimate startup costs, burn rate, first-year revenue, ROI %, and break-even timeline. | SerperDevTool (Live Web Search) | outputs/financial_report.md | initial_investment, projected_roi, break_even_month, first_year_revenue, burn_rate | | 4 | business_validation_agent | Business Validation Specialist: Audits the Financial Report to compute final success probability, risk score, category scores, and a GO / HOLD / NO GO verdict. | SerperDevTool (Live Web Search) | outputs/validation_report.md | market_demand, success_probability, risk_percentage, growth_potential, final_recommendation, demand_score, growth_score, investment_score, risk_score |

🔐 Security & Authentication Architecture (security/)

Managed in security/auth_routes.py:

• Password Hashing: Uses passlib with bcrypt (CryptContext) so plain-text passwords are never stored.
• JWT Session Tokens: Signed with HS256 (JWT_SECRET) with a 7-day expiration (JWT_EXPIRY_DAYS = 7).
• Server-Side Google OAuth 2.0: Validates frontend Google access tokens directly against https://www.googleapis.com/oauth2/v3/userinfo using aiohttp and upserts the user profile in MongoDB.
• Role-Based Admin Access Control (is_admin_check): Restricts sensitive features—such as the real-time WebSocket Live Agent Console (/api/ws/logs)—exclusively to the configured ADMIN_EMAIL.
• Resource Ownership Enforcement: Every report retrieval (GET /api/reports/{id}) and deletion (DELETE /api/reports/{id}) verifies doc["user_id"] == current_user["_id"] before returning data.

⚙️ Backend API & Database Architecture (backend/)

MongoDB Atlas Collections (backend/db.py)

Connected asynchronously via motor.motor_asyncio.AsyncIOMotorClient using certifi TLS certificates to the startup_intelligence database:

1. users: Stores email/Google accounts, hashed passwords, profile metadata (mobile, company, designation, country, timezone), and avatar URLs.
2. reports: Stores full multi-agent analyses, raw markdown outputs, parsed JSON chart metrics (parsed), viability scores, and project metadata (PRJ-XXXXXX).
3. agent_outputs: Logs individual agent executions (trend, discovery, financial, validation) per user.
4. activities: Tracks user audit events (e.g., "Analysis completed: FinTech") for the dashboard activity feed.

REST & WebSocket API Endpoints

| Method | Endpoint | Auth | Description | | :--- | :--- | :--- | :--- | | POST | /api/auth/signup | Public | Register a new user with email & password | | POST | /api/auth/login | Public | Authenticate user and return signed JWT | | POST | /api/auth/google | Public | Verify Google OAuth access token & return JWT | | GET | /api/auth/me | Bearer JWT | Fetch current user profile & admin status | | PUT | /api/auth/profile | Bearer JWT | Update user profile details & upload avatar image | | POST | /api/trend | Bearer JWT | Trigger Agent 1 (trend_crew) & save output | | POST | /api/discovery | Bearer JWT | Trigger Agent 2 (business_crew) with field & budget | | POST | /api/financial | Bearer JWT | Trigger Agent 3 (financial_crew) with business_report | | POST | /api/validation | Bearer JWT | Trigger Agent 4 (validation_crew) with financial_report | | POST | /api/reports | Bearer JWT | Parse all 4 agent outputs via chart_parser.py & save report | | GET | /api/reports | Bearer JWT | List all saved reports for the authenticated user | | GET | /api/reports/{id} | Bearer JWT | Retrieve full report (raw markdown + parsed chart metrics) | | DELETE| /api/reports/{id} | Bearer JWT | Delete a specific report owned by the user | | GET | /api/stats | Bearer JWT | Aggregated KPIs (total analyses, average score, high-risk count, recent activity) | | WS | /api/ws/logs?token=| Admin JWT | Real-time stdout/stderr stream from CrewAI execution (WSStreamLogger) | | GET | /api/models | Public | Query local LM Studio /v1/models endpoint (if active) | | GET | /api/health | Public | Service health check |

💻 Frontend Features (frontend/)

Built with React 19, Vite 8, Tailwind CSS v4, Recharts, Framer Motion, and Lucide Icons:

• Executive Dashboard (Dashboard.jsx): Displays aggregated metrics (Total Analyses, Average Viability Score, High-Risk Ideas Flagged) and recent activity logs.
• New Analysis & Agent Workflow (NewAnalysis.jsx, AgentWorkflow.jsx): Step-by-step interactive pipeline triggering Trend -> Discovery -> Financial -> Validation agents with progress tracking.
• Interactive Report Visualizations (Report.jsx, ReportCharts.jsx): Renders parsed JSON metrics into interactive charts alongside formatted Markdown reports.
• Real-Time Admin Console (LiveConsole.jsx): Streams live CrewAI agent thought processes over WebSockets (/api/ws/logs).
• Run History (RunHistory.jsx): Search, view, and manage past startup validation reports.
• Profile & Theme Management (ProfileSettings.jsx, ThemeContext.jsx): Avatar upload, company/timezone settings, and dark/light theme switching.

🛠️ Setup, Installation & Running Locally

1. Environment Variables

Create a .env file in the root directory (use security/.env.example as a reference):

GROQ_API_KEY=your_groq_api_key
SERPER_API_KEY=your_serper_api_key
MONGODB_URL=mongodb+srv://<user>:<password>@<cluster>.mongodb.net/?retryWrites=true&w=majority
JWT_SECRET=your_super_secret_jwt_key
[email protected]

2. Install Backend Dependencies

Ensure Python >=3.10, <3.14 is installed:

pip install uv
crewai install
pip install fastapi uvicorn motor pymongo[srv] certifi python-dotenv passlib[bcrypt] python-jose[cryptography] python-multipart aiohttp requests

3. Run the FastAPI Backend Server

From the project root:

python backend/api.py

The API will start at http://localhost:8000 (Swagger docs at http://localhost:8000/docs).

4. Run the React Frontend

Open a second terminal and run:

cd frontend
npm install
npm run dev

The web application will be available at http://localhost:5173.

5. Run the CrewAI Pipeline via CLI (Optional)

To run the 4-agent pipeline directly in the terminal and generate reports in outputs/:

python tests/test_agent.py
# or
crewai run