Mar 2023 – Oct 2023 · 7 mos · Bengaluru, Karnataka, India · On-site

Mar 2021 – Dec 2021 · 9 mos · Bengaluru, Karnataka, India

Mar 2020 – Mar 2020 · 0 mo

• This system suggests possible questions that user might have when he/she visits the customer support portal. Every user journey in the app is recorded and sorted according to time. Each user journey corresponds to a specific event so based on user journey a score is given to each event.
• This score decreases exponentially for past events, thus decreasing importance of past events.
• When the user visits the customer support portal, he/she is shown the best three questions related to his/her user journey based on scores given to each event.

Sep 2019 – Mar 2021 · 1 yr 6 mos

• Currently here at okCredit, I am working on chatbot to aid the customer support team.
• 1. We started with dialogflow, which was supporting basic requirements.
• 2. For more complex queries we made our own algorithm to find out patterns in chat history. Once we found patterns we try to match with incoming queries and serve results.
• 3. Now we are moving towards more complex queries. We are vectorizing previous chats and for each incoming query we are trying to find cosine similarity and euclidean distance between vectors to match best possible answers.
• I will update here the rest soon.

Jun 2019 – Aug 2019 · 2 mos · Noida Area, India

• Currently I am working on prediction of age and gender through users' listening pattern. I am trying neural networks for this. Currently I am working on tuning genetic algorithms for neural network, specifically NEAT to predict age and gender of users.

May 2017 – May 2019 · 2 yrs · Noida Area, India

• I am currently working in the Gaana team. Currently I am in search and recommendation team, working on NLP algorithm for entity recognition.
• I work on voice UI for gaana. I have developed my own ranking and searching algorithm to categorize words from the input query, as to which they belong among artist, track, movie, mood, playlist,language and other. The NLP algorithm initially processes the live data from gaana database and maintains its own data structure, so that while matching later, it takes the shortest time possible to find the best category match. To reduce spelling errors, I have implemented a custom spell corrector which corrects spelling and then processes the input.
• My algorithm has been implemented in python and hosted using gunicorn and nginx. Apart from this entity recognition task my algorithm also used lyrics database to avoid null results.

May 2016 – Jul 2016 · 2 mos · Bengaluru Area, India

• Worked on android app as a backend developer. Along with this, in a competition in Flipkart for interns, built a google chrome extension to manage tabs, and won 3rd position for it.