sojharo
commented
7 years ago - I am assuming the there is a confidence factor in API.ai to the question he answers. Some questions would be at 100% if the ask the exact question, some would be at 90% and on some, the bot has no clue or has less confidence. Is this a correct statement? If can we define say if the confidence factor is below 40% then say something like: "Sorry I do understand, please contact xxxx or visit of website....."
Yes, it is a correct statement. Bot is working on rule-based method if question asked falls in certain rule category bot gives the answer correctly. Our bot is not using NLP techniques. These are very sophisticated learning methods and are offered by Microsoft as a solution. Yes, we can show the email address or website in the answer if bot doesn't understand the question.
- On a periodic basis, what is available from retraining perspective. How is the data made available to the trainer on what got answered and what did not? Is it a UI or there is also a data extract. Maybe this is better answered in a demo.
Api.ai doesn't support retraining up till now. So, by using the API our human trainer can see what was the question asked and what was the answer given by the bot. We would have to extract data in regards of this.
- Can we take our instance of API.Ai and move it to our own server. We will probably not do this but the customer was asking that want the flexibility to do that. So just answer it from the perspective of feasibility and how much the work would be.
We can create similar UI on our own dashboard which would mimic the behaviour of API.AI dashboard. Customer would use our UI to train the bot and we in turn would call the API.Ai rest endpoints. We can't take out the rule based logic from API.AI as it is hidden inside their logic and we don't know how API.AI trains the bot behind the scenes.
- I think the current chatbot is not tied to CloudKibo account, please steps I have to do to open and attach this Chatbot there. Just send me the link or we can do over go to the meeting. I want to set it up so multiple people can access it.
There are two bots. One is created by my Google Account and other is created by Zarmeen's google account. The current one that we are using is from my account. We can only create account on API.AI using google. We can import and export bots. We can do it over the go to meeting.
- They want to access to the trainer. We had done the design to provide the trainer. What is the high-level estimate to do this work? Can we give them access to the trainer we are using?
We are using the API.AI dashboard as our trainer. We would have to design the trainer dashboard on our own server. There is a design document for this. Zarmeen might need to review it again and then she would be able to let us know how much work would be required on kiboengage dashboard.
- Is API.Ai multi-tenant or we need to implement a separate instance for each customer? I think the answer is a separate instance. So let's assume we have a chat bot of retail and 90% of the rules are common and 10% is unique. How do we keep 90% common? Is there a modular approach? I just need education
We can create multiple bots per single account. They are called agents on API.AI. They are not sharing rules between agents. If we create rules for one agent we cannot use them for other agents. Agents are defined separately.
zarmeen92
jekram
I had a demo with Vantors on the ChatBot and album dullah, everything went well.
There were a number of question and action items.
These are ALL discussion items. We will later open sperate task to this work where required. Please answer these then we would have a meeting to discuss this.
I am assuming the there is a confidence factor in API.ai to the question he answers. Some questions would be at 100% if the ask the exact question, some would be at 90% and on some, the bot has no clue or has less confidence. Is this a correct statement? If can we define say if the confidence factor is below 40% then say something like: "Sorry I do understand, please contact xxxx or visit of website....."
On a periodic basis, what is available from retraining perspective. How is the data made available to the trainer on what got answered and what did not? Is it a UI or there is also a data extract. Maybe this is better answered in a demo.
Can we take our instance of API.Ai and move it to our own server. We will probably not do this but the customer was asking that want the flexibility to do that. So just answer it from the perspective of feasibility and how much the work would be.
I think the current chatbot is not tied to CloudKibo account, please steps I have to do to open and attach this Chatbot there. Just send me the link or we can do over go to the meeting. I want to set it up so multiple people can access it.
They want to access to the trainer. We had done the design to provide the trainer. What is the high-level estimate to do this work? Can we give them access to the trainer we are using?
Is API.Ai multi-tenant or we need to implement a separate instance for each customer? I think the answer is a separate instance. So let's assume we have a chat bot of retail and 90% of the rules are common and 10% is unique. How do we keep 90% common? Is there a modular approach? I just need education
Please answer these and then we have a meeting to discuss.
@zarmeen92 @dayemsiddiqui