Amazon Q - AIGC Assistant

AI & LLMsGenAI Applications

Topic: Amazon Q - AIGC Assistant

Presenter: Kai

Additional Resources:


System Design Presentation - Amazon Q

Join Us on Wechat

Subscribe to Our YouTube channel https://commitway.com/eventyoutube

https://www.amazon.com/q/

Most search and chat products access only public information

Amazon Q:

access information internal to a company: ERP, CRM, ticket system, customer service, etc.

Summarizes answers across different resources

Amazon Q demo

Users can create multiple chatbots. Each chatbot can specialize in one task

Application set up. Service role creation.

Adding data source

upload document

add data sources

Data source: How to add data source (Sharepoint)

Data source: Adding secret for authentication

Data source: Adding retrieval

Data source: VPC (virtual private cloud)

Data source: Sync scope, sync mode, sync schedule

Sync now

  • Access is secure: only customer can access; amazon employees cannot

Access data through chat interface.

User can look up the original source through link.

Can also upload file on the fly and query it.

Well built security, privacy and quality

When Q doesn’t know the answer

A setting is different between two Q instances.

Turning off access control: it exposes all documents to all employees

Turning on access control: access depends on access control

Setting up Q with web crawler

Setting up access control

Setting up proxy

Sync domain, subdomain

Using web interface

Implementation details.

No fine tuning based on customer data.

No need for training.

Can deal with new data. It does not retrain the model.

Kendra vs Q? Kendra is a separate service

Kendra is an enterprise search

Kendra vs Q is like Bing vs chatGPT

[private network access] - can set up. Usually through VPC

[how to login to Q? How does this map to sharepoint? Re access control]

Can login with same email or username

[does the crawler display who is crawling?]

Not sure

Can use Cloud watch to show it

Respect robots.txt

[external website?]

It’s supported but recommended.

[Can I separate no information vs information not accessible?]

Now need to use cloud watch to display it.

[combine multiple data sources?]

Consolidate the relevant document

[website with login?]

Some are supported.

====

[cached data]

[can external customers use Q?]

[can Q be embedded on website?]

[does the crawler display who is crawling?]

[external website? website with login?]

[combine multiple data sources?]

[how big can the document be?]

[Retrieval augmented generation]

===