Manas Jain

I am working as a Associate ML Scientist at Wadhwani AI . At Wadhwani AI, I am focussing on projects on AI for Social Good. Before this, I have worked as a Research Engineer at Hilabs in the Natural Language Processing Team. At Hilabs, I spent most of my time developing a product which can perform Medical Code Prediction from physician notes. Recently I also was selected to attend Research Week with Google AI, 2022 in the Natural Language Understanding track (Was among the 50 people in the APAC region). Before this, I spent four wonderful years as an undergraduate student at Indian Institute of Technology (IIT), Bombay in Mumbai, India. My major field of study was in Civil Engineering , and on top of that I completed double minors in Computer Science (CSE Dept, IIT Bombay) and the Centre of Machine Intelligence and Data Science (CMInDS).

I was fortunate enough to collaborate with India's best AI researchers during my undergrad at IIT Bombay. I have been working closely with Prof. Pushpak Bhattacharyya and Prof. Sriparna Saha at the IIT Bombay CFILT Lab and IIT Patna AI-NLP-ML Group on my Bachelor Thesis. The project was an industry collaboration with LG Soft India for a real-time problem related to Natural Language Generation (NLG). The problem was to generate human-like responses based on user queries in the domain of electronics products like washing machines, refrigerator etc. This solution was supposed to be used in the conversational AI agent like chatbot or voice assistant. Also I worked with Prof. Soumen Chakraborty on a semester long research project on Multilingual Relation Classification where I fine-tuned multilingual language models for the task of entity relation classification.

I also spent an awesome time working at Daikin Technology and Innovation Center, Osaka, Japan as an AI research intern working on activity recognition in videos using audio transcripts. Previously, I have also worked at Prodigal and Samespace as an AI research intern.

I was also member of the Self Driving Car(SeDriCa) student team (Innovation Cell, IIT Bombay) developing a fully autonomous self-driving car customized for Indian road conditions. I was part of the path planning subsystem and implemented algorithms to find the shortest path in real-time.

My hobbies include playing sports like Table Tennis, Lawn Tennis, cricket & basketball.

Email  /  CV  /  Github  /  LinkedIn  /  Twitter

• [February 2023]  Moved to the Silicon valley & AI capital of India - Bengaluru
• [December 2022]  Starting working as a Associate ML Scientist at Wadhwani AI . Really excited to work in the AI for Social Good domain. Will be moving to Bengaluru soon.
• [February 8 to February 11, 2022]  Attending Research Week with Google. My track is Natural Language Understanding (NLU). I was one of 50 people selected in this track from India and Singapore (APAC). Please feel free to connect with me on socials organised on gather town app. Had interesting discussion with Ankur Parikh, Shashi Narayan and many more during the week.
• [January 2022]  I have been shortlisted to participate in Research Week with Google scheduled from Feb 8 to Feb 11. Really excited to know about the state of art of a variety of ML Techniques and hear from great AI researchers.
• [August 2021]  Virtual Reality based Convocation - Graduated from IIT Bombay with double minors. Check out my VR here
• [July 2021]  Started my first job as a Data Scientist at Hilabs
• [June 2021]  Successfully defended my Bachelor thesis. Writing my first research paper
• [May 2021]  Final semester done, AI & DS minor completed (Major courses of the sem - ASR, Optimization in ML, Intro to ML, seminar, BTP-2, R&D course). SPI - 9.60/10
• [March 2021]  I'll be a Teaching Assistant for the course on Linear Algebra for freshmen (MA 106)
• [Jan 2021]  Excited to study a course about climate change (CM 402 Earth`s Climate: Past, Present and Future)
• [Apr 2020]  Will be interning at Daikin Technology and Innovation Center, Osaka, Japan and Prodigal virtually!!!
• [May 2019]  I'll be interning at Samespace, Mumbai
• [July 2018]  Started my Computer Science minor. Studying Data Structures and Algorithms this semester
• [July 2017]  Started my journey at IIT Bombay as an undergraduate

At Wadhwani AI, I will be leading the ML vertical of the project on Integrated Agriculture News Monitoring, an AI/ML - based surveillance system for the Indian Ministry of Agriculture to track incidences of interest (Pest infestations and Crop diseases) at regular intervals. I am also closely working to build a Conversational AI agent to help automate Farmer Call Centre which will be an AI/ML enabled scalable response system to make the existing Kisan Kall Centre Management System more efficient and use the digitized call data for building a better solution.

US based healthcare-focused artificial Intelligence solution for reducing financial losses caused by data errors. Extensively worked on the problem of medical code prediction used in the health insurance industry from clinical notes (unstructured data consisting of medical domain language, e.g., discharge summary). I developed the end to end pipeline to find evidence (for better explainability) of ICD10 procedure and diagnosis codes within physician notes at scale. This work is currently part of Hilabs product suite and is being used to find DRG upcoding in claims data and improve the HEDIS ratings of US health plans. Also worked on computer vision problem of setting up the OCR pipeline of multiple scanned images. Improved the google tesseract OCR accuracy by implementing line segmentation algorithms as a precursor to the OCR engine.

I am broadly interested in Natural Language Processing, Speech Recognition, Computer Vision and Deep Learning. My specific research interests include Text Generation, Conversational AI, Abstractive Summarization, explainability and fairness in AI.

Question Answering systems these days typically use template-based language generation. Though adequate for a domain-specific task, these systems are too restrictive and predefined for domain-independent systems. We propose a system that outputs a full-length answer given a question and the extracted factoid answer (short spans such as named entities) as the input. Our system uses constituency and dependency parse trees of questions. A transformer-based Grammar Error Correction model GECToR (2020), is used as a postprocessing step for better fluency.

The overall aim of the project was to build a model which could learn to classify relations in low-resource language from data in high resource language (English) using pre-trained multilingual language models. We fine-tuned MuRIL (Multilingual Representations for Indian Languages), BERT, and RoBERTa on a custom-made Relation Classification dataset (English) and benchmarked performances on a manually created test set.

Worked as an intern under the theme of Advanced Digital Engineering Technique utilizing AI technology in the team DigiNavi, ICT group at Technology and Innovation Centre, Osaka. Implemented NLP-based video tagging, captioning, summarizing audio transcripts using TF-IDF based entity count algorithm, BERT QnA model, and Distill BERT extractive summarizer model, respectively.

Worked as a summer intern at Prodigal Technologies (Y-Combinator backed). Developed a sentiment scoring model on collection call conversation having agent and borrower transcripts using unsupervised lexicon and rule based sentiment analysis model VADER.

Part of a team of over 25+ students to build India’s 1st Level 5 autonomy carin a 5-tier challenge (prize money- $1 million). Amongst the top 11 teams out of 259 to be awarded Mahindra e2o electric vehicle for further research and development. I was mainly involved in Motion Planning and Decision Making subsystems of the vehicle where I implemented bidirectional RRT Star algorithm between two points in a 2D-Grid having multiple obstacles of random shape using the concept of Costmap in ROS C++.

Worked as a summer intern at Samespace. Developed Deep Learning models using BiLSTM, GLoVe embeddings in Keras and Tensorflow, Worked on conversational call transcripts to predict tags for sales acceleration. Also trained models to predict sentiments of conversation and profanity detection.

We implement a transformer based architecture for cross-lingual summarization using shared encoder-decoder architecture. The traditional approaches involves sequential application of machine translation followed by machine summarization or machine summarization followed by machine translation. However, such a method would result in tremendous usage of memory and would result in error propagation. We also create novel dataset for the purpose of cross lingual summarization from english to hindi. We also have open sourced the code for a novel state of the art architecture for cross-lingual summarization whose implementation is currently unavailable.

Our project involves generating a facial animation video that is automatically synthesized based on speech signals using an end-to-end model. The input to the system is a still image of a person and a speech audio signal. The output would be a video in which the person’s face is animated in sync with the audio. Modified ImaGINator generator to take speech signal as input and implemented 3 Discriminator. Trained the model on GRID Dataset.

Studied Grid search, random search, hyperband, Bayesian Optimization, genetic algorithms to perform HPO. Experimented the above HPO algorithms for regression, classification task on 4 ML models - RF, SVM, KNN, ANN.

Trained KNN, SVM and FFNN on GTZAN dataset having 58 features; achieved 89% accuracy using FFNN. Implemented a Convolutional Neural Network architecture and trained it on Mel-Spectrograms extracted from raw audio using Librosa.