How to Build a Responsible AI Chatbot Using Open Technologies 7 Sections35 Lessons5 QuizzesLifetime

Curriculum

7 Sections
35 Lessons
Lifetime

Expand all sectionsCollapse all sections

Introduction
2 Items
- 1.1
  The CABot Story
- 1.2
  How to Do This Course
Module 1 - Understanding AI
9 Items
Large Language Models (LLMs) generate language, not truth. Explore tokens, prompts, context windows and hallucinations, then examine why Retrieval-Augmented Generation (RAG) grounds responses in trusted knowledge.
Module 2 - Building the Knowledge Base
9 Items
Understand how to transform a digital resource (e.g. a book) into semantic knowledge through chunking, embeddings and vector search. Learn why CABot uses ChromaDB and metadata rather than retraining models.
Module 3 - Choosing Your Architecture
5 Items
Compare hosted APIs, local models, vector databases and orchestration frameworks. Rather than recommending one stack, this module helps you choose technologies appropriate to your own community.
Module 4 - Building the Chatbot
8 Items
In this module, we follow a learner's question from a browser prompt to asynchronous RAG orchestration. We aim to understand asynchronous architecture and browser-first deployment.
Module 5 - Responsible & Multilingual AI
5 Items
Responsible AI begins with trustworthy knowledge, not simply powerful models. Explore transparency, source attribution, multilingual retrieval, privacy, accessibility and environmentally conscious AI.
Course Conclusion & Next Steps
2 Items
- 7.1
  Lessons learned
- 7.2
  Recap: Practical workflow for building a responsible chatbot

Why a task queue?

LLM responses may take seconds or longer. If the web server waits while doing all the work itself, many users can make it unresponsive.

Celery acts as a job coordinator:

Flask accepts the question quickly.
Celery places it in a work queue.
A worker handles retrieval and generation.
The browser asks, ‘Is it ready yet?’ every 1.5 seconds.

This pattern improves responsiveness, although a larger production system might use streaming or other real-time delivery methods instead of polling.

Retrieval plus generation

The central RAG loop is:

The repository searches the Climate Academy collection and, if available, an optional encyclopedia collection. It merges results, sorts them by vector distance, applies a threshold.

Flask

How to Build a Responsible AI Chatbot Using Open Technologies 7 Sections35 Lessons5 QuizzesLifetime

Curriculum

Why a task queue?

Retrieval plus generation

Modal title