AI WhatsApp Voice Transaction Chatbot
Budget: $30 – $250 USD
I need a Python-based AI chatbot that runs inside WhatsApp and understands both voice notes and typed messages. Using OpenAI’s API (or a comparable LLM) the bot should:
• Transcribe and interpret incoming voice notes,
• Accept text-based entries of sales, expenses or any other transaction,
• Categorise every transaction automatically,
• Store all data securely so it can be queried later,
• Correct entries when users send follow-up fixes,
• Push a concise daily business summary each night and an automated weekly report every Monday,
• Return full transaction history on request.
You are free to wire the logic through the official WhatsApp Business Cloud API, Twilio, or another compliant gateway, as long as the end-user experience stays inside a single WhatsApp chat. The model should be lightweight enough to run as a SaaS micro-service (FastAPI preferred) and expose clean endpoints for future front-end or web-app expansion.
Acceptance criteria
– Voice note accuracy ≥ 90 % in English.
– Categorisation precision ≥ 85 % across at least 10 custom categories.
– Daily and weekly summaries delivered on schedule for two consecutive weeks in staging.
– All data stored in a relational DB with basic CRUD endpoints documented in Swagger.
Source code in a private Git repo, a one-click deploy script (Docker or Render), plus a brief README are the final deliverables.
• Transcribe and interpret incoming voice notes,
• Accept text-based entries of sales, expenses or any other transaction,
• Categorise every transaction automatically,
• Store all data securely so it can be queried later,
• Correct entries when users send follow-up fixes,
• Push a concise daily business summary each night and an automated weekly report every Monday,
• Return full transaction history on request.
You are free to wire the logic through the official WhatsApp Business Cloud API, Twilio, or another compliant gateway, as long as the end-user experience stays inside a single WhatsApp chat. The model should be lightweight enough to run as a SaaS micro-service (FastAPI preferred) and expose clean endpoints for future front-end or web-app expansion.
Acceptance criteria
– Voice note accuracy ≥ 90 % in English.
– Categorisation precision ≥ 85 % across at least 10 custom categories.
– Daily and weekly summaries delivered on schedule for two consecutive weeks in staging.
– All data stored in a relational DB with basic CRUD endpoints documented in Swagger.
Source code in a private Git repo, a one-click deploy script (Docker or Render), plus a brief README are the final deliverables.