Inspiration

Growing up, I watched my father struggle with manual billing for his industrial and agricultural business because small business owners often lack the capital, infrastructure, or desk setups for complex enterprise software. Years later, I wanted to build a solution to alleviate this systemic administrative burden for the broader small-business community.

What it does

The pipeline automatically parses the unstructured data, updates a central database, and triggers a clean, professional invoice back to the client. Instead of querying a database, users can simply ask the chat agent, "What are my sales this month?" or "Who is my top client?" to get instant, executive-level insights.

How we built it

I built an end-to-end Agentic AI system using n8n, Gemini’s multi-modal API, and Supabase. I chose WhatsApp as the interface because it eliminated user hesitation. I designed logic loops where a user could just snap a photo of a handwritten multilingual bill or send a voice note.

Challenges we ran into

  1. Complex bills with many items were being cut off mid-sentence because of a low token ceiling, and the AI occasionally included markdown backticks or conversational text that prevented the data from being parsed.
  2. The "Phone Match" logic incorrectly sent empty results down the "Success" path because n8n's "Always Output Data" creates a dummy object that looks "true" to basic filters.
  3. Could not create a WhatsApp Business Account, so used an imitation API called Evolution API.

Accomplishments that we're proud of

The solution successfully digitized manual tracking workflows, reducing the administrative time and manual effort for these resource-constrained businesses by 80%.

What we learned

Building an application from scratch with minimal and ambiguous requirements.

What's next for AI Billing Automation

Need to hand this application over to vendors and merchants and test that the application is working in real-world scenarios appropriately without any hesitation from them.

Built With

Share this project:

Updates