Tips and tricks on how to create AI customer support assistants and sales AI assistants

We are currently developing advanced AI customer support assistants.

Anyone else here?

I think it would be beneficial to exchange some experiences in this field.

1 Like

Would be interested in this as well.

I know a number of people in Ecom who would love to automated some or all of their support tickets.

1 Like

I highly recommend RASA, as you can finetune it and specify intents and actions based on the user input. It does just that, but it does it extremely well. Speaking of he open source version. The benefit is that you won’t ever produce hallucinations and only return predefined answers. It can’t just mix up things and provide false information. A huge flaw in most LLMs today, as I found.

If you still want to bet on an easier to manage horse, I’d recommend you look into Copilot Studio. You can apparently finetune it very specifically (still waiting on my access, since you have to be a registered business to get an account) but from what I’ve seen in tutorials and such, it’s … very impressive!

Here’s a demo for what it can do:

https://web.powerva.microsoft.com/tryit

1 Like

Thank you for sharing.
I was briefly checked the website and read few articles, but this not what i need.

We are building our own highly customizable system and we plan deployment in a wide range of uses, customer support is only the first option because there is a great demand for this service.

I already saw few hybrid systems (old way “tree structure” bots blended with ai) and its not what we are looking. We are building 100% AI system.

Have a great day! :slight_smile:

1 Like

Hm! I’m curious now:

You are refering to bots blended with AI, are you addressing RASA or Copilot or both? Sure you can use RASA for predefined answers in a more traditional way (which is preferred by many due to its simplicity and the control you have over the responses it delivers) but you can also create your own actions, forwarding requests sorted by defined intents toa generative AI which then replies based on the action you defined.

Basically turning a simple bot into the router of a complex construct that can not only give precise responses, but actually do things, too. E.g. run scripts that can feed back into it. Really opens up the possibilities right there.

I’m not trying to convince you, but I’m interested in learning more. If you don’t mind, would you be able to provide a more specific reason as to why this is not an option? Or what a 100% AI system looks like? I can’t even imagine that, to be honest.

Again, just my curiosity! If you don’t want to spend that time, that’s fine. I guess I want to understand what you’re doing :smile:

Hello Wes :slight_smile:

I would like to say thank you, because at the beginning of this idea was your video a few months ago. Thank you.

I know some of great system architects and developers and I brought them together to work on this project. We are currently developing our system based on GPT and we are creating multi-agent architecture of experts to achieve the best possible results when working with high amount of data (currently user manuals). We are focusing to create highly customizable solution and we approach each client individually.

The project is still in early stage, but a lot of work is done and we are currently cooperating with first clients to obtain a data from real use cases. We already building first solution.

If you know anybody who want to participate in our project in exchange to early access and much better price, i really like to hear about them. If you are interested in it, i can tell you more in PM.

Thank you for your YT channel, which is a great and fun way how to obtain new informations in this field! Im big fan of your channel!

Lukas

1 Like

We are betting on AI system. Actually its not 100% AI, as there is an application which we are creating and the structure of this app is important part of system.

System of predefined answers can be reliable solution but it requires a lot of time to prepare and set for client. We believe that 100% AI assistants are the future.

Thank you for your interest. :slight_smile:

1 Like

I don’t want to offend anyone, however, I really have a hard time when I contact customer support and can’t understand the person on the other end due to their accent. I am hoping AI puts an end to that. I live in Texas and the first thing you usually hear is press 1 for English or 2 for Spanish. It should prompt you for your choice of language and then continue in that language. With AI, it should be able to handle the majority of all common languages. Can someone please make that happen?

Yes, exactly. This is where AI can really help. Because working at customer service is really damaging human mental health. I have no problem to talk/write with AI on customer service, if i don’t have wait for connection or push milion+ buttons to get what i need to get. It will save me a time and also mine mental health as well…

Hello

I myself am currently trying to discover the possibilities of a simular structure. In the POC my goal is to set up a custom chat application that handles as a personal HR/company advisor to an employee. Guiding them in their journey in the company and facilitating the procedures.

I started messing around with Voiceflow and find it a great application to take control of the conversation through NLP but also provide the ability to have a conversation without the fixed structure. Worth the time to explore and test a conversational structure.
That said, the platform has disadvantages like high tokens counts and price, limited language options, etc.

I am now diverting into a full backend application that just uses whatsapp business api as an UI. The main reason is that i think that the UI is not the most essential part for the POC right now.
Next up would be embedding the knowledge base and storing it in a vector database on a cloud server. As of now i am relying on the openai assistant that embeds it on it’s own and gets stored at openai. Not realy attractive for b2b partners because of storing data at third party companies.

If there’s a way i could help or to brainstorm, you can reach me at tom.l@lutosolutions.com.

1 Like

Hello Tom

And thank you for feedback! I checked the voiceflow, looks capable but we are working on our own app solution. I believe that custom solution will be much more attractive for clients than “Do it yourself” approach services.

We currently working on similar think - embeddings. OpenAI assistant works good with pdfs and other sources, problem is when there is much more pdfs to know. One assistant is simply not handle it very well (yet), thats why we step into multi-agent (specialist) space.

We currently working with products XMLs and user manuals, which are open data sources, so storing it in third party is not such a problem. We want to step into open-source once it will be reliable enough, because privacy is highly valued among customers.

Thank you for contact but i want to keep discussion open in case that somebody want to join and brainstorm together. :slight_smile:

Success with that! Sounds like a solide solution to offer customers.
If capital is present, watsonX from IBM might also be interesting for training models, hosting data lakes and monitoring the performance of the LLM’s. The entire platform can get quite expansive. It’s aimed at enterprise level.

Glad to hear you’ll keep the discussion open. Would like to hear other people’s thoughts aswel.

1 Like

Thank you!

And thank you for tip for watsonX!

Hey all, check out this video - using 3 platforms it shows you how to get this going easily:

Relevance ai
Pipedream

Enjoy!

(Limited to only 2 links)

hehe seems we are competitors I am currently working on something very similar :smile:

1 Like

No, we’re not! I am simply trying out things for the sake of feeding my curiosity. But I’d be happy to read more about what you’re working on, now that you mentioned it :grin:

1 Like

I dont think we are direct competitors, because im located in different continent (EU). Can you share more details about your project? This thread should be about sharing tips…