AI Backed Personal Assistant Chatbot (That Speaks Burmese)

Myanmar’s number of internet users has risen from 2 million (2014) to more than 39 million (2016). Number of SIM cards in circulation has risen by almost 400 percent (43.72 million SIM cards sold). This translates to 89.38% mobile penetration. But there’s a big language and technology barrier for people in Myanmar, my home country. I started to think further about it and wondered what if there’s something I could to help narrow this gap?


My vision for this product is to simply create Alexa that speaks Burmese. I was still at Amazon and started to learn more about how I could use Alexa platform and add Burmese language support. Alexa at the moment couldn’t integrate with other languages except English. The team has been trying with languages like Japanese and German which already has strong libraries, dev communities, or similar grammatical structures. In the case of Burmese language, it doesn’t have any of those. So even early on, I realized that I will have to build my own AI system. So I started writing the most simplest information architecture down.

My initial contender for translating Burmese to English and English to Burmese was Google Translate. I started trying the sample phrases that I would support to see how trustworthy Google Translate is and it was much better than as I have initially hoped. This is when I was confident enought to reach out to my friend Aung Htet who has excessive knowlege in information architecture and Burmese language. He liked my idea, my draft IA, and joined me immediately.

I believe this AI assistant needs to be funny and smart and should be a man. After all, I do think it’s a little bit sexist to give all AI assistant female names. I decided to use the name U Paw Oo since it’s a name of a comedic character who was a senior advisor to Burmese kings for several decades. From time to time, he made fun of his kings, too. It fits perfectly and that’s how U Paw Oo was born.


Since this is my own product, I serve as product & UX lead, and my friend Aung Htet serves as tech lead. Additionally, I also helped him in training the AI system.


1. Choosing the right platform:
Burmese were late in joining the global internet community and thus, there are significant userbase for Skype, Viber, WhatsApp, Google Hangout, and Facebook Messenger.

2. Finding the right content:
As both Aung and I have been out of Myanmar for around 10 years, we both are not familiar with daily life of a person living in Myanmar. So what are the type of services they would want to know?

3. Data consumption:
Every developing country has users with this concern. 

4. Lack of adoption for unicode standard font:
Zawgyi font is the most popular font used by Burmese all over the world. The only problem is that it is not up to Unicode standard. There are groups advocating for Unicode standard Burmese font but adoption is still significantly low.

5. Are my users ready for AI assistant?:
Both Alexa and Google Assistant are not in Burmese yet. iOS and Mac marketshare is less than 1% so Siri is also out of the question. 


50% of total mobile users in Myanmar mainly use the device for IM and VOIP. Myanmar has a total of 14-16 million facebook users and 90% of them uses messenger platform. And they use facebook daily according to 2017 data. This shows that Facebook Messenger has the biggest market share and was an still is the most advanced of them all. Besides, we built our chatbot in a way where we can deploy on any instant messaging platform we want. In fact, we can even create our own stand alone app one day if we want. So we decided to use Messenger.


Although we have a huge list of services we want to add to our chatbot, we looked into what type of information people in Myanmar consume and found that local news updates, weather updates, and soccer news are the highest sought for new categories. I also wanted to put some services that would help business owners and people with family members in foreign countries. 

So the list of services I decided to add were News, Weather, Forex, International Time, Soccer Results, and Soccer Upcoming Matches.


I somehow already had an answer for this and it was to use text as much as I can to reduce data consumption. But the main challenge to me as a product designer is how to balance between data consumption and the user experience of the product.

I started looking into the news sources where most Burmese visited, the local Facebook page with the higheast number of likes and follows, and personal interviews with friends and family. Something I found out which also supported my thinking is that although images are more engaging, it’s still ok to just read the title to make a decision on whether or not to read the article or watch the video. I felt confident to just put titles of the news articles, youtube videos, and just soccer team names.

Entertainment News Articles
Video News from Youtube Channels
Soccer Match Scores
Upcoming Soccer Matches


Since the beginning, my hyphothesis is that news will be the most requested service and I was right. And it is also one of the cheapest development from our end. We basically used Feedly to connect hand vetted and curated news sources by me to our chatbot. But in order to create a perfect UX for mobile, we made the requested news category to show only 3 at a time with show more button right below the results. User can also move from one category to another just by clicking on the shortcut buttons below. 

News categories: Entertainment, Sports, Financial, Politics, Local, Lifestyle, Videos
Showing 3 stories at a time with show more button below. 
Other categories are also listed below for quick switch.


As I have mentioned above, weather is the second most sought information by Burmese. It is a little bit conflicting with the personal interviews we did with our friends and family since they mentioned that weather is not necessarily useful for Myanmar since weather is predictable in Myanmar. Maybe it is because all of our friends live in Yangon and they all drive for commute. As most Burmese commute using public transportation, It could be useful for our users. Plus, it's a low hanging fruit in terms of implementation.

When asked, the bot ask for which city the user would want to know with suggestions listed below
In the weather result text, we mentioned date, location, degree, and cloud formation. We also added other cities as suggestions


As a Burmese living abroad with family still in Myanmar, one issue I always ran into was to figure out what time it is there. This is also true for me and my co-founder since he lives in Sydney, Australia. It's also a pretty low hanging fruit for us to do proof of concept.

For time, we ask for which city the user would like to know with an instruction text to type any city name if not shown below.
After showing the time, we add other highly requested cities from our users with an option to restart to the main menu.


If you run a business in Myanmar, you need to know USD, SGD, Chinese Yuan, India rupees, and Thai bahts. If you have family members out side of Myanmar, you need to know that. Supporting foreign exchange rate is a must and again not that hard to implement.

In this service, using a full country name would make the text string too long and using currency code is also something users will remember to use. So I thought about using an emoji of flags and started assigning them to country names. It created a very pleasant experience and interface for users.

It starts with the 10 most commonly asked currencies for locals with an option to show the rest
When tapped, the bot will show all the rest of the currencies 
We are also mentioning that the rates are from the Central Bank of Myanmar. This helps the creditbility of the rates


Soccer is the sport of the country. All European major leagues have tens of millions of viewership every week. We believed that providing LiveScore service to our users without the need to load a webpage with ads, and on demand will help reduce down a huge data consumption. Besides it will also increase the usage for us. For first release we divided the soccer info into two parts, past results and upcoming matches. It's not ideal but at the time it would be an easy way to build out the foundation.

Premiere League, Champions League, English Championship, La Liga, Bundesliga, Serie A, & FA Cup
Showing results from the last week
Showing upcoming matches

The structure is where if a user request soccer results, the bot would ask back which league. Based on the users second input, the bot would reply back with information. We knew that it is still not the ideal experience but it aligned well with the information structure we're getting from the API. It's also the same for getting upcoming matches and we believed it is enough for alpha version.


Since chatbots are new to Burmese users, I have to think of every use cases and try to make sure there are not deadends in all the flows as much as I can.

The first question we ask at the beginning is “which font do you want to use?” in both Unicode and non-Unicode format
For all the appreciations the user has shown by saying thank you to thumbs up, we always reply something like no worries, or you are welcome
A restart keyword is always at the end of all the options, and once clicked, we reshown all the main categories with font switcher at the front


We knew that the bot's current ux is not really at an ideal stage. There are a lot of improvements that we could do to current services we are supporting. We plan to support more features every quarter and current structure would create a bad ux if we add more.

That's why we already have plans to start introducing card UI and group similar services into categories. This way users will be able to access many more services at a higher level.


My intention for this product was never to create a chatbot with shortcut buttons. I wanted The users to interact with it just like they would interact with a human. A long the way, I've added human like behaviors into the product. There are personal questions users can ask upo like his age, his work, how he feels about the user, his significant other, and the list goes on.

If U Paw Oo is asked about his age, he would sometimes say “I won’t tell you” and sometimes “In books, I have been alive since the last dynasty. But I just got online not too long ago"
If U Paw Oo is asked whether he has a significant other, he would sometimes say “I won’t tell you”, and sometimes say “I am single. Hm Hm Hm” in a very pervy way
If U Paw Oo is asked about his job, he would sometimes say “just this and that” or “I am ready to do anything you asked me to do, buddy"
If asked about his name, he would say “This is the last time telling you, it’s U Paw Oo” or “Are you kidding me? I am U Paw Oo"
If asked whether he likes you, he would answer “I will tell you when we meet outside"
If asked what he thinks about you, he would say “Haven’t met you outside but I already know you are pretty”, or “You are my master so you must be very intelligent"

We put in Natural Language Processing inside the backend to make sure that the bot would understand almost anything our users would ask in the future. Even now, our users can ask anything at any level. What that means is, you might be inside asking for soccer scores but without saying restart or going to other categories, you can ask specific news, or weather or local time of a specific city.

Asking Yangon local time while inside soccer score service
Asking Yangon weather while inside local time serivce
Asking about Aung San Suu Kyi news while inside weather service
Asking about US$ exchange rate while inside news service


All these steps we've taken is to reach to our ultimate goal in the future where U Paw Oo can be a voice Assistant. You can imagine people with language and technical barriers can use their smartphones just like you and I are using right now but only cooler. 

I will keep posting of the new improvements and features for U Paw Oo whenever I can.

For confidentiality reasons, I have taken out the actual numbers from all the companies I have worked for and have put percentage. These are all just my views on the project I worked on and does not represent the views on the organizations.

© 2018