PayFit - SEO Copilot
AI-powered SaaS built in a hackathon to automate SEO content optimization at scale
The Problem
PayFit produces SEO content continuously — HR articles, legal updates, accounting guides. The problem: finding the right topic, sourcing information, writing, checking legal compliance, getting validation. A cycle taking 3 hours per article, with a real risk of error on legal points. The hackathon brief: imagine and build a tool that automates this workflow end to end.
Approach
Solo on the project — architecture design, N8N workflow construction, Dust agent configuration and full technical spec writing in 4 days.
I first mapped PayFit's existing workflow: what sources they consult, how they identify trending topics, how legal validation happens. This let me design an automation that fits their reality rather than a generic tool.
N8N workflows automatically retrieve recent articles from HR/payroll competitors, Google PAA (People Also Ask) on target topics, and new French laws in labor and accounting — to identify high-potential SEO topics before competitors.
Dust agents handle two distinct tasks: article writing (with retrieved sources as context) and legal fact-checking (via an up-to-date compliance database that verifies the content contains no inaccuracies on legal points).
The solution was technically functional at the end of the hackathon — operational N8N workflows, configured agents, complete spec for the Next.js interface development.
Technical Details
N8N as central orchestrator: workflows chain competitive monitoring, PAA retrieval, legal monitoring and Dust agent triggering — all configurable without touching code for non-technical teams.
Specialized Dust agents: a writing agent with sources as context, a legal agent with access to a compliance database. Separating into two distinct agents allows the legal database to be updated independently without impacting the writing workflow.
Multi-tenant architecture designed for Supabase: each editorial team has its own space, article history and monitored topics — designed for simultaneous adoption by multiple PayFit teams.
Complete SPECS.md technical spec: data model, workflow schema, API endpoints, agent structure — deliverable ready for a development team without ambiguity.
Challenges & Solutions
The solution was technically too advanced for PayFit's internal hosting — they didn't see themselves operating an N8N + Dust infrastructure independently. This is an important lesson: a good technical solution must also be one the client can own and maintain. I factor this into how I design internal tools since then.
In 4 days solo, aggressive prioritization was required. I chose to deliver functional N8N workflows rather than a polished interface — a technical jury can evaluate a running workflow, not a mockup.
Results & Impact
- Functional N8N workflows delivered in 4 days — competitive monitoring, PAA retrieval, legal monitoring, operational Dust agents.
- Article creation cycle: 3 hours (find topic, source, write, validate) → 5 minutes (validate only).
- Directly reusable architecture — I'm applying the same logic to build an equivalent SEO tool at Antalis (KPP Group).
- Best technical solution of the hackathon according to jury feedback, with clear feedback on operational inadequacy for PayFit.
Architecture
Scheduled or manual trigger → N8N orchestrates: (1) competitor article scraping + Google PAA + new laws → (2) high-potential topic identification → (3) Dust writing agent with sources as context → (4) Dust legal verification agent → (5) validated article ready to publish in Next.js interface → history stored in Supabase.
What I Learned
This hackathon taught me a lesson I wouldn't have learned otherwise: the best technical solution isn't always the right product solution. Delivering something the client can't operate alone means the work isn't finished. Since then, I start every project by asking 'who will maintain this?' before choosing the stack.