Overview
AstroVerse is an AI-powered business strategy platform that transforms a raw startup idea into 11 different investor-ready PDF documents in minutes. Built in a 22-hour hackathon sprint (1st Place out of 30+ teams), it runs as 4 independent services: a Next.js 15.5.4 frontend (React 19.1, shadcn/ui, Tailwind 4, Turbopack) on port 3000, a FastAPI authentication backend (SQLModel + JWT/bcrypt, port 8001), an n8n self-hosted workflow engine (port 5678) orchestrating 11 specialized AI agents, and a FastAPI Finalizer Service (WeasyPrint 61.2 + Jinja2, port 8000) with 11 dedicated PDF generation endpoints — one per business document type.
The Problem
Aspiring entrepreneurs in Tunisia need professional business plans to secure funding, but can't afford consultants. Existing AI tools produce generic text that lacks market-specific data, financial projections, and the visual polish investors expect. There's no tool that generates a complete, multi-section, investment-ready business document suite from a single idea description.
Solution
Designed an 11-agent n8n workflow where each agent specializes in one business domain: Incubator (multi-concept comparison), Market Analyst (competitive landscape + sizing), Strategist (Business Model Canvas), Brand Architect (naming + identity + guidelines), Product Manager (user personas + stories + roadmap), Legal Assistant (IP + compliance + regulatory), CMO (SEO + content calendar + social), CRO (sales funnels + pricing + partnerships), Production Ops (Agile methodology + QA), Growth Strategist (beta testing + metrics + launch coordination), and Business Planner (all-in-one comprehensive plan). Each agent's output chains forward as context to subsequent agents.
Built the Finalizer Service (FastAPI 0.104.1 on port 8000) with 11 dedicated POST endpoints (`/generate-incubator-pdf`, `/generate-market-analysis-pdf`, etc.), each backed by a custom Jinja2 HTML template. WeasyPrint 61.2 renders the templates into print-ready PDFs with branded headers, charts, tables, and professional typography. The repo ships with 8 sample output PDFs (MelodyMuse brand identity, legal compliance, marketing strategy, product strategy, strategic analysis; SyncSonix growth strategy, operations plan, revenue strategy).
Created the FastAPI authentication backend (port 8001, FastAPI 0.118, SQLModel 0.0.25 + SQLAlchemy 2.0, JWT via python-jose, bcrypt 5.0) with user registration/login, saved idea management (`/prompts/` CRUD), and a `/workflows/` endpoint that triggers the n8n webhook to launch the full agent pipeline.
Built the Next.js 15.5.4 frontend (React 19.1, TypeScript, Tailwind 4, Turbopack dev server) using shadcn/ui components (Radix primitives + class-variance-authority), Lucide icons, and dark mode support. The dashboard shows real-time agent progress during generation.
Architecture
Next.js 15.5.4 frontend (port 3000) submits the business idea to the FastAPI auth backend (port 8001), which triggers an n8n webhook (port 5678). n8n orchestrates the 11-agent pipeline — each agent is an n8n node calling the LLM with a specialized system prompt and context from previous agents. Agent outputs are forwarded to the Finalizer Service (FastAPI port 8000), which renders each document as a professional PDF using Jinja2 templates + WeasyPrint. The frontend polls for progress and provides download links when PDFs complete. SQLModel (SQLite via `astroflow.db`) stores users, saved prompts, and workflow metadata.
Tech Stack
Next.js 15.5.4 + React 19.1
shadcn/ui (Radix + CVA), Tailwind 4, Turbopack, Lucide icons, dark mode, real-time agent progress tracking
FastAPI Auth Backend (:8001)
FastAPI 0.118 + SQLModel 0.0.25 + SQLAlchemy 2.0 — JWT auth (python-jose), bcrypt, user/prompt/workflow CRUD
n8n Workflow Engine (:5678)
Self-hosted n8n — 11-agent pipeline with context chaining, webhook triggers, and conditional logic
Finalizer Service (:8000)
FastAPI 0.104.1 + WeasyPrint 61.2 + Jinja2 3.1.2 — 11 POST endpoints generating branded, print-ready PDF documents
11 AI Agents
Incubator, Market Analyst, Strategist, Brand Architect, Product Manager, Legal Assistant, CMO, CRO, Production Ops, Growth Strategist, Business Planner
Results
Won against 30+ teams in a 22-hour sprint
Each specialized in one business domain — from market analysis to legal compliance
Professional Jinja2 + WeasyPrint documents with charts, tables, and branded headers
Frontend (:3000) + Auth (:8001) + n8n (:5678) + Finalizer (:8000)
MelodyMuse (5 docs) + SyncSonix (3 docs) — shipped in the repo as proof of quality
Achievement Gallery
Moments that made it all worth it
Achievement 01
🏆 1st Place — AstroPreneur Hackathon, November 2025
Key Takeaways
n8n's visual workflow paradigm is a game-changer for multi-agent systems — visual debugging and hot-reloading saved us hours during the hackathon. Being able to see data flow between all 11 agents in real-time was the difference between finishing and not.
Context chaining between agents is critical: without it, the financial projections contradicted the market analysis. We solved this by passing summarized context forward through the n8n pipeline so each agent builds on the previous one's conclusions.
WeasyPrint is underrated for PDF generation — it supports modern CSS (flexbox, grid, custom fonts), which means we designed reports in the browser with Jinja2 templates and rendered them identically as print-ready PDFs. No LaTeX, no Puppeteer.
Splitting the system into 4 services (frontend, auth backend, n8n, finalizer) during a 22-hour sprint sounds risky, but it meant 5 team members could work in parallel from hour one without merge conflicts or blocking.