Business Document Generator for LLM Consumption

Summary

I built a Business Document Generator that automates the creation of detailed reports by orchestrating AI calls in n8n, managing inputs in Airtable, and invoking specialized tools for tasks like data scraping and charting. This system produces human-readable documents and structured outputs for downstream AI processes, cutting manual prep time in half and ensuring consistency across reports.

Client’s Brief

My clients needed a way to generate comprehensive business documents—combining narrative, data tables, and visualizations—without manual assembly. They wanted a workflow that:

Pulls structured parameters from Airtable
Uses AI to draft and refine narrative sections
Calls external tools for tasks like web scraping or chart generation
Outputs documents usable by both humans and other AI agents
The solution had to scale easily, allowing new features to be added via template updates.

Tech Stack & Approach

n8n – Orchestrates workflow, handles branching, and invokes tool-calling logic
Airtable – Central hub for templates, input parameters, and result storage
OpenAI o4-mini – Powers reliable content generation and refinement
Google Gemini Flash – Optional fast drafts, swapped in for non-critical steps
Custom Tool Calls – Integrates scraping and charting services via webhooks
Webhooks & APIs – Connects orchestrator to external data sources and visualization tools

Building the Solution

I started by designing an n8n workflow that retrieves document templates and input fields from Airtable. The orchestrator then:

Outlines & Drafts – Splits into AI steps for outlining and drafting narrative sections.
Model Selection – Routes non-critical drafts through Gemini Flash for speed, while sending final refinement requests to OpenAI o4-mini for accuracy.
Tool Invocation – Calls custom scraping and chart-generation services, then writes results back to Airtable.

I iterated on prompt structure to balance clarity and flexibility, and built error-handling nodes to catch failed API calls without halting the entire workflow.

Impact

50%+ time savings in document prep, from ~4 hours down to under 2 hours per report
90%+ accuracy in data-driven sections via automated validation checks
Plug-and-play scalability: added new scraping and visualization features through template updates
Consistent quality across documents, reducing revision rounds by 30%

What I Learned

Keep workflows simple: fewer branches lead to greater reliability under load.
Choose models strategically: speed-optimized drafts vs. accuracy-focused refinements.
Low-code tools excel for rapid prototyping—Airtable and n8n cut development time dramatically.

Next Steps

Add a web-scraping module to auto-fetch industry benchmarks.
Embed dynamic charts directly into final documents for richer insights.
Implement user-feedback metrics to refine prompt strategies and template structures.

Closing CTA

If you could automate any part of your reporting workflow with AI, what would you tackle first? Let me know below!