AI Email Support Assistant
Budget / Salary$250–750
TypeFreelance project
LocationRemote
Posted1 hour ago
I need an AI-driven platform that reads incoming customer emails about orders, returns, or account issues and instantly drafts clear, accurate replies my support team can approve with a single click. The first release must work entirely over email; live chat or social channels can wait until this core workflow feels rock-solid.
Key requirements
• Secure IMAP/SMTP or Gmail API connection to pull new messages and push the drafted response back into the same thread.
• Intent detection and entity extraction so the AI understands order numbers, return reasons, account details, etc.
• Draft generation powered by a large-language-model pipeline (OpenAI, Claude, or a custom fine-tuned model—open to your recommendation) with guardrails to keep tone on brand and avoid hallucinations.
• Human-in-the-loop dashboard where staff can review, edit, approve, or reject each draft; actions should sync back to the customer in real time once approved.
• Logging and simple analytics so I can track accuracy, average approval time, and the percentage of emails fully resolved by the first draft.
Acceptance criteria
1. At least 90 % of test emails about standard order, return, or account questions produce a usable draft that needs minimal editing.
2. Latency from email receipt to draft display is under 20 seconds.
3. All customer data stays encrypted in transit and at rest.
4. Solution deploys on my preferred cloud (AWS) with clear setup docs and hand-off.
Provide a short outline of your proposed architecture, the NLP/LLM stack you’d choose, and any similar projects you’ve completed so I can gauge fit.
Key requirements
• Secure IMAP/SMTP or Gmail API connection to pull new messages and push the drafted response back into the same thread.
• Intent detection and entity extraction so the AI understands order numbers, return reasons, account details, etc.
• Draft generation powered by a large-language-model pipeline (OpenAI, Claude, or a custom fine-tuned model—open to your recommendation) with guardrails to keep tone on brand and avoid hallucinations.
• Human-in-the-loop dashboard where staff can review, edit, approve, or reject each draft; actions should sync back to the customer in real time once approved.
• Logging and simple analytics so I can track accuracy, average approval time, and the percentage of emails fully resolved by the first draft.
Acceptance criteria
1. At least 90 % of test emails about standard order, return, or account questions produce a usable draft that needs minimal editing.
2. Latency from email receipt to draft display is under 20 seconds.
3. All customer data stays encrypted in transit and at rest.
4. Solution deploys on my preferred cloud (AWS) with clear setup docs and hand-off.
Provide a short outline of your proposed architecture, the NLP/LLM stack you’d choose, and any similar projects you’ve completed so I can gauge fit.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.