Context
Rural producers often record expenses through WhatsApp messages, photos of receipts, and voice notes. The result is scattered information in mixed formats that is hard to consolidate into reliable financial control.
Problem
Manually recording and categorizing these expenses is time-consuming, error-prone, and makes it hard to build a solid financial base for decision-making. Every lost receipt or forgotten entry distorts the real picture of the business.
Solution
I built an AI-powered finance assistant that receives messages directly over WhatsApp — text, audio, or photo. It uses OCR and transcription to extract the content, natural-language processing (an LLM) to interpret amounts, dates, and categories, and a smart validation flow that asks the user to confirm whenever something is ambiguous.
I chose to orchestrate the whole flow in n8n, wired to the WhatsApp API and OpenAI, with entries persisted to a structured store (Google Sheets) via webhooks. This keeps costs low, removes the need for a dedicated app, and meets users where they already are — on WhatsApp. The assistant also learns from previously confirmed classifications, improving its accuracy over time.
Result
The outcome is a low-cost, instantly adoptable solution: the producer records financial activity using nothing but WhatsApp, and unstructured information becomes organized, analysis-ready data. The MVP already handles text, audio, and images end to end, eliminating manual entry of financial records.