Digit Geek
U.S. Navy photo by Mass Communication Specialist 3rd Class Joshua M. Tolbert/Released
Digit Geek > Recent Articles > ALT > Backtalk – Animesh Samuel of Light Information Systems on natural language processing

Backtalk – Animesh Samuel of Light Information Systems on natural language processing

We have a conversation with Animesh Samuel on using natural language processing for process automation.

Digit: Can you explain to us a bit about what NLPBots does?  

Samuel: I like to give the example of this sentence, “I shot an elephant in my pajamas, one morning.” The question is that in this sentence, I being myself, what am I likely to shoot the elephant with? Did I shoot the elephant with a camera, with a camera on a mobile phone or with a gun? Who is wearing pajamas, is it the elephant or me? The idea is that if we can teach the machine that, then we can solve various use cases. Chatbots is just one of them. NLPBots is a horizontal AI NLP platform, which can mix both structured and unstructured data and solve different kinds of processes, or communication related challenges. For example, we can automate the entire hiring process for when you have the KRA or JD, all the way to the first interview. The bots themselves will read the resume, rank the resume depending on how close they are to the JD, it will communicate with the candidate automatically by accessing the phone number and email, it will then conduct screening with the candidate and the candidate can also ask questions about the profile. Then, if both the candidate and the company are satisfied, it can schedule the first interview. The entire process, end to end can be automated using our technology. There are employment engagement chatbots, there are marketing insights for which Tata Communications, our client won the marketing excellence award, where the bots will read through various dynamic resources and collect lots of data on the prospects. So, before any employee goes to make a sale or a pitch, they have all the information about the clients. All this is done on our AI NLP platform, which basically can do this kind of a process automation.

Digit: What kind of languages and input mechanisms do you support?

Samuel: We work with English, Hinglish and Hindi right now, but we have also had a roadmap for other Indian languages and some international languages as well.

The input mechanism can be anything, it can be a WhatsApp chat, it can be an HTML page, it can be Telegram, or it can be a voice bot as well, such as Amazon’s Alexa. We can process any kind of voice or text input.

Digit: Indic languages have traditionally been low resource; do you use monolingual or bilingual data sets?

Samuel: We have more than 40 of our own proprietary algorithms which are trained on neural networks. For example, if it is a customer related thing, mostly it is just one particular language. In terms of corporate data, we don’t have enough in other languages. We translate it to English, then process it, and then translate it back, in cases of low resource local languages, where there is not enough data available. We use both monolingual and bilingual data sets, but that depends on the problem. If we have enough data for training, then we can use monolingual data sets as well.

Digit: What kind of data do you use to train the neural networks?

Samuel: The data for training normally belongs to the customer, it is not our data. For example, if you are Mahindra, you have seventy policies for six different employee bands, all that is your data. I will train on your data.

We have a platform where you can load all your structured and unstructured data. The structured data is loaded by linking to APIs, the unstructured data is loaded by just feeding the documents, maybe connecting a SharePoint or feeding all the documents separately. Then it trains on both those documents and the API using our algorithms and neural networks.

Digit: Are there any differences in the way you handle spoken or written Hindi for example?

Samuel: All our processes are in text only. If anything spoken also comes, we convert that into text first then work on it. All NLP AI in the world is text only. Voice only AI is used for authentication and stuff like that and is not used for getting answers.

Digit: Can you tell us a bit more about the team? How much of your technologies are developed in house?

Samuel: We are a fifty-member engineering team, based out of Pune. Most of them are doing their masters, have done their masters, and some are doing their PHDs also, in machine learning or artificial intelligence. All our technologies are developed in house and our algorithms are proprietary, and our neural networks are proprietary. Our applications can be loaded on any cloud of your choice, or any premises. We are not bound by either the front end, or interfaces, and on the back end also it can be any kind of structured or unstructured data. We are an exciting company at an exciting time. When we started off, NLP was neuro-linguistic programming, it was not even natural language processing. We used to tell people you need to think of it that way. Today, everyone knows it and understands it, and its potential is huge. As a platform and product company, we are the only one in India in the B2B space. We are looking at the next level, and we have the best clients also, including Tata Communications, Edelweiss, Piramal, Mahindra and exciting startups such as Oyo. In various use cases, we automate different processes and conversations also. We are not just a chatbot company, we are a process automation company that does chatbots as well.   

Aditya Madanapalle

Aditya Madanapalle

An avid reader of the magazine, who ended up working at Digit after studying journalism, game design and ancient runes. When not egging on arguments in the Digit forum, can be found playing with LEGO sets meant for 9 to 14-year-olds.