Build AI Customer Support Chatbots
From FAQ bot to a grounded assistant with handoff to humans.
Taught by Rohan Maharjan · ML Engineer
What you'll learn
- Design a support assistant's scope, intents and tone from real ticket data
- Ground answers in your own help-centre content with retrieval-augmented generation and citations
- Route intents to tool calls that look up orders, check policies and create tickets
- Implement confidence-based escalation and a clean handoff packet for human agents
- Deploy the same assistant to a web widget and the WhatsApp Business Cloud API
- Measure deflection, CSAT and answer quality, and run a weekly improvement loop
About this course
Build AI Customer Support Chatbots is a hands-on course for developers, product engineers and technical founders who want to ship a support assistant that answers accurately, takes real actions and knows when to hand over to a person. A customer support chatbot is the most common first AI project for a business, and most of them fail on the same things: hallucinated answers, no route to a human and no way to measure whether they help. You will build a Python assistant step by step: design intents and tone from real ticket data, ground answers in your help centre with retrieval-augmented generation, add tool calls for order lookups and refunds, implement escalation and human handoff, and connect the bot to a web widget and the WhatsApp Business Cloud API. The final lessons cover guardrails, privacy and measuring deflection and CSAT. You should be comfortable with Python and basic web APIs; no machine learning background is required.
Course content
8 lessons- 1Anatomy of a Support Bot That Actually HelpsThe three levels of support automation, the components every good assistant shares, and what to build first.Preview
- 2Designing the Conversation: Intents, Scope and ToneMine ticket data for intents, define what is in and out of scope, and write a system prompt that sets tone and boundaries.
- 3Grounding Answers in Your Knowledge BaseChunk your help centre, embed it into a vector store, retrieve the right passages and force the model to cite them.
- 4Intent Routing and Structured ActionsClassify what the customer wants, and let the model call tools for order lookups and ticket creation with strict schemas.
- 5Escalation and Human HandoffDetect when to escalate, hand over a complete context packet, and make the transition feel seamless to the customer.
- 6Channels: Web Widget and WhatsAppExpose the assistant through an HTTP API, embed a web chat widget, and connect the WhatsApp Business Cloud API webhook.
- 7Guardrails, Privacy and Failure ModesStop the assistant from leaking data, making promises it cannot keep, or being manipulated by prompt injection.
- 8Measuring Deflection, CSAT and Continuous ImprovementDefine the metrics that show whether the bot helps, compute them from logs, and run a weekly loop that makes it better.