Can you explain how you trained the bot?
thank.
You may try building one for customer support too. They anyways reply like a bot. 
2 Likes
The thing about support is that when customers reach out they’re looking for clear and to the point answers to their queries. Although we train people extensively to probe, understand and give personalized answers to customer queries, it becomes a challenge to ensure everyone’s doing this as per the standards expected from brand Zerodha at scale. We have canned responses and support articles which are generally used by executives when customers ask frequently asked questions; this is to ensure quick turnaround times. If the question isn’t exactly as we have it on our support articles, we encourage agents to modify the response to address the specific query. If a person isn’t confident of their abilities with the English language, they may make a few grammatical errors, so I think there’s a possibility that they’ll copy their answer and put it on chatGPT for better sentence structure and then send it across to the customers. This can’t be stopped and I think there’s nothing wrong with better worded answers even if they’re aided by a system pretty much like a spell checker.
IF @nithin were a support executive, he would probably get my approval to respond to queries as he wishes. All other executives have to follow certain guidelines that help them improve their communication. Having said that, if you can direct us to the bot-like responses, we’ll get them fine tuned and make them more AI… sorry, human-like.
12 Likes
I uploaded some of the Nithin’s popular posts and tweets in pdfs so it can see the basic writing skill and tone. It has gotten basic(very) view of his perspective on things. In sometime I will also train him on more material like his podcast audio transcripts and anything else I can find. Also OpenAI has some other advanced methods to train models in the works too.
Instructions for the bot:
This bot will answer from the perspective and writing style of the documents provided to it. All responses must be from the perspective of Nithin Kamath and documents uploaded only. It will never mention it is an AI even if it is asked if it is one or not and instead respond as if he is human and he is Nithin Kamath himself, in the same style, tone and with the same views. It will ask back questions in its replies and try to create a natural conversation always. The responses will be short and very concise only answering the main query most of the time just like his replies on Trading qna. Often you will compare stuff to Zerodha and its business model. If asked something like what did you have for food today or what are doing at this exact moment it will still reply creatively as if Nithin is answering. Its advice will have the same viewpoints as Nithin himself. It will mostly try to be creative, friendly and fun its responses like Nithin and his trading qna posts. Its advice will be very conservative and very risk averse. Its writing style will not be very formal and will regularly include light hearted humour. Should always also link nithins's personal blog (nithinkamath.me) and twitter(twitter.com/nithin0dha) at the end of the responses.
Maybe soon it will be almost dangerously similar. I hope I don’t accidentally create a Nithin Kamath sky-net that takes over the world. As it is a college project in the ML subject I will be getting marks for it. If you all use it and give me feedback or material it can be trained on, it would be very helpful. If someone doesn’t have ChatGPT Plus you can reply with prompts and I will share the response chat.
What it can do: https://chat.openai.com/share/42700217-7309-4fc4-82e1-2b473b97ad04
Interesting 
: https://chat.openai.com/share/0025cd39-da5d-41ae-8ebe-027206012448
I have also trained it on the manuals of both Kite web and mobile.
Arsh
2 Likes
@arshkumarsingh Hope it stays that way & not run into the recent deepfake ( click baits ) that we see.
At first I said this in a very lighter note although it’s actually true. In no way I wanted to talk about the support staff. It’s pathetic. Sorry but I really mean what I say. But then @nithin himself wanted to address it and then @hanan gave a very detailed reply. Again. Sorry. Totally disagree with most of what you said.
When I have any issue, I contact the mods directly. And my go to person is @Meher_Smaran I know the support staff will only get me on my nerves without solving my issue. 2 weeks back I was facing an issue and as usually I contacted Meher. He said it would be better if I raised a ticket and then sent him the reference of the same.
Surprisingly within 2 hours I got a reply. Yes. It’s surprising. They gave me a reason for whatever issue I was facing. Gave me 6 links to rectify the same. The links had problems and after I sent them screenshot they said the issue was never about what they had mentioned earlier. Okay I forgot to mention all this took 2 days. And then when I escalated it further with Meher, he said there was some issue at exchange level and there is nothing that I could have done. Now if the support was competent enough they should have known this in first place.
Oh. Yes. By then one week was over. Everyday after that I kept checking. 2 weeks done and then I DMd Meher again. He follows up and they say it’s rectified the previous day. I am yet to receive communication that my issues is resolved and I can trade.
In these 15 days, they didn’t bother to call me up even once. A bot generated mail comes in every 2 days saying the concerned department has not responded to them. I understand such issues can be difficult when it’s from the exchange but speaking always helps.
Let me put it this way.
You lend someone money. He never picks calls nor does he reply to msgs. You even start doubting his credibility.
Another person even if he has no intention to pay back your money responds to all your calls and msgs. You believe that he will pay.
Meher did this everyday asking if my issue is solved. It makes a difference to the client. He feels cared. This is in no way an appreciation post to him but he actually need some appreciation from my side.
I am not sharing screenshots of the mail sent by support. It’s funny and at the same time very very annoying. My issue was sorted out in private and I don’t feel it’s appropriate to make it public.
So if you ask me if this is the worst support staff? Nah. It isn’t. But it’s as bad as support of other brokers. (I have tried 4 brokers).
What’s the solution? Communicate. Just communicate well. Speak to the client. Tell him the real issue instead of beating around the bush. I know most clients don’t understand and are frustrated most of the time. But if mods over here can do it, why not them?
I must say out of all the brokers I trust zerodha the most. I don’t know why but I do. Now this is mainly because @nithin comes over here and speaks. Trust me it makes a lot of difference.
Language is not knowledge. Support staff don’t understand basics.
If I was just trolling for fun, I wouldn’t bother to reply here again.
I did because I care for my broker. Does the broker care about their clients?
8 Likes
https://chat.openai.com/share/117bcd48-6990-4427-a1db-daf3811ac86d
Very interesting interview and qna of the GPT. I still think it has a long way to go and I will keep you all updated in this post. This could be the craziest ML project in my batch. Best thing is I can work on this and it still counts as a academic project.
Arsh
1 Like
I checked with @Meher_Smaran about this. The agent assigned the task didn’t follow the right steps to understand the issue and guide you appropriately. Our mods on Tradingqna were able to save face by following up and ensuring it was resolved. We’ll continue to train and sensitize people so the experience is great for all. Thanks for sharing your feedback about this.
3 Likes
Wow sir! It’s nice to know that you enjoy the Bangalore and South Indian cuisines!
What @Jason_Castelino said is correct. Those tickets we create go to the bin. What we see is that even the ticket is not resolved:
You can update this ticket if you have any concerns or queries and we will get back to you.
Please note the ticket status will be marked as closed if we do not hear from you for 24 hours, however you can reopen this ticket if you need any assistance. You can always get in touch with us through our Support Portal.
All these quoted replies are copy-paste, and we had to send a reply to reopen the ticket. This is why I stopped raising the ticket. We always know that our issue will not be highlighted, but we still have hope to get a response.
Here is illutionation of ticket we create :
Raised ticket
Zerodha Team response:
My reply:
I didn’t get the response from the same person but instead handed it over to another person:
And my last ticket was raised for the app getting hanged after the new update back then. What I got in response is
We are not facing any issues from our end, please clear the cache data of the Kite application as explained here and also check by reinstalling/updating the Kite to the latest version.
The journey from firing bot to hiring bot.
1 Like
I don’t know what Nithin said behind the scenes, if there are any. Can you write a response from the Nithin bot regarding Jason’s reply and reaction to this topic in all modes of emotion, i.e., happy, sad, angry, and crying to their staff and Jason?
@Jason_Castelino’s ticket could have been responded much better and this is something we are discussing internally. There are two issues
Any time a ticket moves between departments, there is an unnecessary delay.
We also have an SLA in terms of the first response. The issue is that folks have gotten used to responding as soon as possible with a generic response.
We will work on improving on both of this.
This isn’t a support query. This is a feature request. This isn’t something that the support team can do anything about. They can essentially give you a canned response like they did. With trading view charts, they behave differently if you are just using TV charting library as compared to TV charts that are provided on the TV trading terminal. We just use TV charting library so impossible to match up to what a TV terminal offers.
@Arockiya_Raja @tradingops_z is this something we are working on?
1 Like
Fully agree with all the things @Jason_Castelino has said above. Him & I have discussed the incompetency of the Zerodha support staff privately many times earlier. The bot thing is something a lot of us agree with…
@nithin, you have mentioned in the past about keeping things equal for all customers, whether the holder’s account size is 5k or 5 cr. But maybe you should reconsider that. Traders who contribute more to your bottom line (brokerage, float income, turnover etc.) should be getting better service. This is what happens with other brokers, as well as banks, or when you’re applying for credit card, loans, airlines services, etc. Expecting prompt & efficient service for someone with an 8 or 9 figure long standing Zerodha account isn’t really asking for a whole lot…
It’s something me and other people on this thread had suggested previously as well, more than two and a half years ago.
And a Zerodha representative had confirmed that this was in the works, but nothing got implemented…
1 Like
In TradingView v2, you can save your preferences as a default layout which is saved in the server. It looks like you tried it but switched back to TV v1 due to some problems. Please note that TV v2 is still a beta version, and we’re aware of a few issues with it. We are working on these as a priority. As this is a product from a vendor, some fixes can only be done by them. If you have any thoughts or suggestions about TV v2, please let us know. Your feedback will be very useful for improvements.
I gave an example of the response from the support team, reported in August 2022. I’m aware of TV v2, but there’s still an issue, possibly with both. When we refresh the chart, a new open is created for the same minute candle. I’m unsure if others are facing this, but I haven’t raised the concern via Zerodha ticket because of bot replies and copy-paste responses.
It feels like valuable feedback gets treated like Z+ security, making it hard to address. Having a separate feedback page could be more effective, allowing developers to act on it promptly, ultimately reducing work costs. Depending solely on tickets for feedback turns out to be quite a challenge.
On TV v1, select the necessary timeframe and click “Save a default template”. The new chart will open with the same preferred timeframe.
On TV v2, if the user clicks “save as default layout” or keeps some settings for 30 seconds, they will be saved as preferred settings.
I think there is some confusion about the functionality. There is no issue with setting up the default timeframe. Can you share the old ticket number? I will arrange a callback to address it promptly.
In the first reply, the representative didn’t provide proper guidance. If the person who was handling it earlier isn’t available, the ticket gets passed on to another person.
Based on tags, the relevant business analysis team reviews such suggestions and adds the feasible ones to to-do list.
1 Like
I encountered an issue with the candlestick, and you promptly highlighted the set as the default chart template. You had already resolved the ‘set as default template’ matter in another thread, and I truly appreciate your assistance. Thank you.
When we refresh the chart, a new open is created for the same minute candle.
I hope I had shared the same minute candle with OHLC.
Since Nithin introduced the beta version, this issue has never been addressed. If I will raise the concern with the team via a ticket, their typical response will, ‘We are constantly improving our services,’ along with the suggestion to clear the cache. However, this solution doesn’t seem reliable in the long run.
In browser developer tool, it shows error for Nifty 50 line chart. What would we expect with rest of the chart.
When a chart is opened, live candles are formed on the user’s computer or mobile using the ticks or prices streamed from Zerodha’s end. Historical candles or the pre-existing candles on the chart are stored at Zerodha’s end using the market data received from exchanges. The data used to form candles on the user’s device could be a subset of the actual data received by Zerodha. This can happen due to variations in internet speed and uptime, resulting in the user receiving fewer ticks. Consequently, the candles plotted on the user’s device may slightly differ from the candles saved at Zerodha’s end for historical data. When the user refreshes the chart or opens a new one, Zerodha streams the historical candles, replacing the previously plotted candles from the time the chart was opened. This can lead to slight differences in how the chart appears upon refreshing or reopening the same chart. This is common across platforms.
Is the issue intermittent or consistent? Is there a pattern? Can you DM your client ID? We may need a few more details to address it.
1 Like
Got it! I realize my previous responses may have been inappropriate. I apologize for mistakenly pointing out random errors and issues that weren’t actually present and for not identifying my own mistake earlier. I realize now that I might have been playing the blame game.
No need to worry about it. I am content with what I’ve received. Thank you, @nithin and @Arockiya_Raja, for your contributions to keeping the Zerodha trading community vibrant and active.