AI-Driven Support Automation Platform
Budget / SalaryHourly project
TypeFreelance project
LocationRemote
Posted2 hours ago
I want a Python-based platform that automatically handles incoming customer messages on two fronts: email and major social-media / messaging channels. The core purpose is to ease support workloads by letting an AI engine
• instantly answer routine questions on its own, and
• prepare draft replies for anything more complex so staff can review and send.
Key expectations
• Written entirely in Python, leveraging common NLP / LLM tooling (e.g., OpenAI API, LangChain or similar).
• Simple connection layer for at least Gmail/IMAP plus one social channel (Facebook Messenger, WhatsApp Business API, or comparable) with room to add more endpoints later.
• Central knowledge-base that the model consults before responding; plain-text or Markdown files are fine for now.
• Confidence threshold setting: if the model is unsure, it stores a draft instead of sending.
• Dashboard or CLI output that logs every interaction, showing whether the reply was auto-sent or queued for review.
Deliverables
1. Clean, well-commented source code in a public or private Git repo.
2. Setup instructions so I can run the service on my own server (Docker preferred).
3. A short README explaining how to extend the knowledge-base and add new channels.
Acceptance criteria
• Messages arriving via both chosen channels get classified within one second.
• At least 80 % of common FAQ-style queries are answered automatically with factual accuracy pulled from the knowledge-base.
• Draft responses clearly separate AI text from any inserted placeholders for human follow-up.
Key expectations
• Written entirely in Python, leveraging common NLP / LLM tooling (e.g., OpenAI API, LangChain or similar).
• Simple connection layer for at least Gmail/IMAP plus one social channel (Facebook Messenger, WhatsApp Business API, or comparable) with room to add more endpoints later.
• Central knowledge-base that the model consults before responding; plain-text or Markdown files are fine for now.
• Confidence threshold setting: if the model is unsure, it stores a draft instead of sending.
• Dashboard or CLI output that logs every interaction, showing whether the reply was auto-sent or queued for review.
Deliverables
1. Clean, well-commented source code in a public or private Git repo.
2. Setup instructions so I can run the service on my own server (Docker preferred).
3. A short README explaining how to extend the knowledge-base and add new channels.
Acceptance criteria
• Messages arriving via both chosen channels get classified within one second.
• At least 80 % of common FAQ-style queries are answered automatically with factual accuracy pulled from the knowledge-base.
• Draft responses clearly separate AI text from any inserted placeholders for human follow-up.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.