About Me

Ashutosh Mishra

I am Ashutosh Mishra. I am from Bhubaneswar, Odisha. I am interested in machine learning and Artificial Intelligence. I love how the ML is changing the world. I aim at contributing as much as I can to these fields. My research interests include Generative Adversarial Networks, Attention models, Sequence to Sequence models and Adversarial Learning. My current research projects are on Deep representation of visual descriptions and LimbicAI which is an AI that induces emotion like fear into an AI. Apart from my interests in AI I have a keen interest in Japanese Culture and Space.I love reading blogs on space and AI in my free time.

My Career


• Extraction of high value analytics from Legal Corpus • Extraction of common provisions, clauses, and data points from Contracts with 90% precision • Building robust ML pipelines with concurrent API end points • Improved in-house API to serve multiple requests faster by 10% • Lessened speed of text-based captcha solving by 25% using SOTA object detection • Improvising fuzzy text matching to be 25% faster on large scale text matching of quoted text between multiple Legal Corpus • Programmed a Case summarization tool of relative chronology of case proceedings using SOTA transformers based models • Developed a first person reasoning model to identify the polarity of speaker • Building, scaling and deploying an ensemble model of transformer and supervised machine learning algorithms to extract information in form of one-class classification • Incorporated a sentence similarity model to find similar text wrt. context of the text using SOTA Natural Language Understanding techniques • Assembled a system to help customers identify in case of force-majeure or renewal of contracts using dependency parsing and word2vec

Jun. 2019
Associate Data Scientist

Kalika Gourmet LLC

Developing backend for deploying a server for day-stock prediction and trading. Predicting long term dependencies for NIFTY using xgboost and ensemble of LSTM and GRU.

Aug. 2018
Data Science Intern

Tino IQ

Financial data modeling.Statistical analysis on long term prediction for hedge-funds like S&P 500 with 92% accuracy.Testing and developing models for stock price prediction.

Jun. 2018
Machine Learning Intern

IntelIndexer LLC

Process economic and finance domain algorithms and methods for feature extraction.Use LSTM,Dynamic Time Wraping, Clustering and scalable methods to build prediction system.Create a GDP prediction system with 5% more accurate than IMF.

Feb. 2018
Data Science Intern

B. Tech

Studied Computer Science Engineering and Applications.

Apr. 2019
Computer Science Graduate

Higher Secondary

Studied Computer Science(12th Class) in DAV Public School, PKT and graduated in the year 2015.

Apr. 2015

My Skills

My Projects

Deep Representation of Visual Descriptions

Generating images(real or abstract) from text descriptions using Attention-GAN and Stack-GAN. This project focuses on text description-to-image conversion through self attention which results in better accuracy.

...   ...


LimbicAI is an AI that induces emotions like fear in AI. This project aims to provide safe AI using self-attention and adversarial learning to remove fake detection, mis-classification and other problems which arises due to noise or tampering with the models from outside.

...   ...

Unique Doodle Generation

Generating new doodle images for different images using Generative Adversarial Networks. Tech - Python , Deep Learning, GANs.

...   ...

Image Caption Generator

Generate Captions for images using VGG16 network. Tech - Python , Deep Learning, torch, CNN.

...   ...


OpenCV application for face, eyes, smile detection. Building MSI/.exe for windows. Tech - Python , OpenCV, cx-freeze.

...   ...


The AI “THE-BOT” is a community driven smart bot which is build using flask and Restful API responsive, which gives you a Weather, Distance and Your_IP.

...   ...


This project is about getting mass results from BPUT using Selenium and BS4.

...   ...

Web Crawler

This app crawls all the links it finds from a website and stores it in the same directory under a manually specified project name. Tech Stack - Python, scrapping, threading.

...   ...