Apr 2022 – Jan 2024 · 1 yr 9 mos · Bangalore Urban, Karnataka, India

• Have developed multiple products at Flipkart working on both Engineering and Data Science.
• Have been leading a team of 12 SDEs for 3+ years.
• Currently leading 4 Engineering teams:
• Vernac powering all the vernacular content on Flipkart through the MT platform which supports Machine Translation, Transliteration and Language detection across 11 Indian languages powered by in-house ML models.
• UGC responsible for soliciting, storing, moderating and serving all Flipkart User Reviews, QnA and Wishlists.
• Voice powering Flipkart’s in-house Voice Search across multiple languages.
• NLP powering all NLP apis for use cases described below.
• Created and led the development of:
• Flipkart’s in house Voice Search in multiple languages which saw excellent user adoption with industry leading ASR and NLP metrics.
• Voice based Grocery shopping and Fashion Shopping Assistants taking ownership of key components such as NLP, Transliteration, Speech Recognition and Dialog Context management. Designed the novel context management system to handle multi turn user queries required for refining product search results.
• Flipkart’s NLP platform for Intent and Entity Recognition. The platform supports defining intents and entities for a domain, labelling training data, predefined entities with canonical representations, automated training, pluggable models and serving of the models through APIs.
• Created NLP ML models for Intent and Entity recognition for Voice Search, Grocery assistant, Fashion Assistant, Search queries and Decision Assistant.
• Created the Language and Acoustic ML models for Voice Search.
• Created Machine transliteration ML models. For this work, have filed a patent for Machine transliteration of Code Mixed Data.
• Did extensive research and experimented with multiple SOTA DL approaches to improve accuracy. Also led the data generation and worked closely with ops teams in coming up with the labelling guidelines and refining them through regular feedback.

Apr 2020 – Apr 2022 · 2 yrs · Bangalore Urban, Karnataka, India

Sep 2018 – Apr 2020 · 1 yr 7 mos · Bangalore Urban, Karnataka, India

Jul 2016 – Sep 2018 · 2 yrs 2 mos · Bangalore Urban, Karnataka, India

• A "Microsoft Accelerator" AI Start Up which got acquired by Flipkart.
• Joined liv.ai as an early stage startup and donned multiple hats - Machine Learning Researcher, ML Engineer as well as Backend Developer.
• Machine Learning Researcher and Engineer:
• Read research papers regularly, applied SOTA architectures on our problem statements as well as created novel DL architectures such as TreeLSTM.
• NLP domain problems:
• Intent and Entity recognition for creating voicebots.
• Transliteration between Indic and Roman scripts.
• G2P (grapheme to phoneme) model for creating a typing tool similar to Google Input tools for transliteration based typing in Indic scripts.
• Neural Machine translation between Hindi and English through Seq2Seq models and transformers.
• Language modelling
• Speech domain problems:
• ASR (Automated Speech Recognition) - Worked in all major components - acoustic model, decoder and language model.
• Language classifier based on Voice for 10 Indian languages.
• Speaker diarization to separate out Agent and User audio clips in single channel telephonic conversations.
• Backend developer:
• Independently designed and created a framework for creating voicebots. Had to don a product manager’s hat to decide the features of the framework. The framework is general enough and can be applied to any domain and had features on par with Google’s Dialog Flow and Microsoft’s Luis. Created voicebots for Flight Booking and Stock trading domains using the framework.
• Created APIs for multiple problem statements such as transliteration, translation, language classifier, speaker diarization.
• Created optimized systems in C++ for transliteration and G2P. Gained deep understanding of GPU architectures to implement highly optimized persistent LSTM architecture for ASR acoustic model. The latency of Liv ASR API reduced by 10X with persistent LSTM enabling it to scale easily with minimal hardware.

Mar 2016 – Jul 2016 · 4 mos · Bengaluru Area, India

• Building progyrus.com to take internet to the real India.
• Progyrus aims to help the internet reach the 300 million Indians who will experience the internet for the first time in the next 5 years. We believe that localization will be the key to tap into the next wave of Indian internet users who will be older, more mobile, more gender equal and more vernacular. We provide human-powered translation service scaled by technology.
• Summarized below is my work experience as a tech co-founder of Progyrus:
• Designed the complete backend including database design and API interface. The tech stack involves Django + PostgreSQL for backend and AngularJS + JQuery for frontend.
• Processed different document formats, and created translation memory to avoid redundant translations and reduce translation cost.
• Used Django REST framework for creating rest APIs and celery for asynchronous tasks. Gained extensive knowledge of localization industry including web and software localization practices.

Jul 2013 – Feb 2016 · 2 yrs 7 mos · Hyderabad, India

• Worked as a software engineer in CDMA protocol team at Qualcomm.
• CDMA protocol stack development, working on layer 3.
• Developed multiple complex CDMA features with involvement in all phases - Requirements Analysis, Design, Coding, Unit Testing and Maintenance. Co-ordinated with multiple teams in deciding the feature design and inter-module interactions.
• Triaged and fixed issues reported from lab/field/operator/customer testing. Provided fixes for many critical issues to Qualcomm customers in a limited timeframe.
• Gained experience in writing high quality code in C meeting the industry standards of efficiency, stability, readability and extensibility.
• Filed 2 patents related to CDMA and Multi-SIM technologies with the US patent office.