Contact person: LLM WhatsApp Sales AI Agent

Phone:Show

Email:Show

Location: New Delhi, India

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"I want to move all day-to-day WhatsApp conversations for sales and support over to an LLM-driven agent. The goal is to replace our current template-based rules with a model that can understand free-form messages, reference live data, and craft responses that feel genuinely human.

Core functions
• Handle customer inquiries, supply detailed product information and give up-to-the-minute order-tracking updates directly in the chat.
• Generate fully customizable replies that improve over time by learning from each customer’s history stored in our CRM.
• Surface personalized offers or reminders by pulling preference signals from past conversations and purchase behavior.

Systems the bot must speak to
• CRM for contact details, deal stages, and conversation history.
• Inventory/fulfillment system to fetch stock levels and shipping status.
• Payment gateway (Stripe or similar) to drop secure pay-now links, confirm transactions, and issue receipts.

Tech expectations
• Use the official WhatsApp Business Cloud API (or Twilio WhatsApp) as the delivery channel.
• Power the conversation with a modern Large Language Model—GPT-4, Claude, Llama 2 or equivalent—wired up through LangChain, Rasa, or a comparable orchestration layer.
• Build the integration layer in Python or Node.js, with clean, well-documented code and webhook endpoints we can host on our existing AWS stack.
• Fail-safes: intent confidence thresholds, automatic hand-off to a human agent, analytics logging and versioned prompt management.

Deliverables
1. End-to-end working WhatsApp agent deployed in our dev environment, then promoted to production.
2. Source code and environment files in a private Git repo.
3. API keys/secrets managed through AWS Parameter Store or Vault.
4. Short video walkthrough and written run-book for our team to fine-tune prompts or swap models later.

Acceptance criteria
• 90 %+ of common inquiries resolved without human intervention over a one-week pilot.
• Personalized product suggestions appear when prior purchases exist in CRM.
• Order-status replies are accurate within 5 seconds of the data held in our inventory DB.
• Payment links open the correct checkout with pre-filled order totals.

If you’ve already shipped something similar on WhatsApp, let’s talk." (client-provided description)