1. Ashutosh Vishnoi
NLP | Deep learning | Machine learning
ashutosh.vishnoi.790@gmail.com +91-7985521330
linkedin.com/in/ashutosh-vishnoi-23a0299a github.com/ashuvishnoi
EDUCATION
07/2016 – 07/2020
B-Tech
IIIT Jabalpur
ECE
WORK EXPERIENCE
06/2019 – 11/2019
Machine learning Internship ( 6- Months )
Unfound.ai
Mumbai
AI based News and Digital Media Company
Worked on Unfound SaaS B2B Product in developing the ML
features (Mostly Rest APIs) and deploy it on aws server for
production purpose.
Worked on some ML features in the product unfound.news
Done some other tasks like aws server setups, Nginx/uwsgi setups,
aws ami creations, to live the api in RapidApi platform etc.
To productionize the ML services on cloud and maintain it with
debugging the issues coming up.
Mentor: Ankur Pandey – https://www.linkedin.com/in/ankurpandey42/
11/2019 – Present
Machine learning Internship (6- Months)
A2J Technologies
Bangalore
Digitalizing law using AI
Worked on preparing vision and analytics features in product
'TDRx'.
Analytics of Company data to find the pattern of rise and falls of
any comapany.
Mentor: Dr Chiranjiv Roy – https://www.linkedin.com/in/chiranjiv/
SKILLS
Natural language processing Deep learning
Machine learning Computer Vision AWS GIT
Python Programming Matlab C Programming IOT
PERSONAL PROJECTS
Distractors Generation
Given question and answer , generate max three distractors( similar
answers).
Cleaned the data and covert it into vectors using glove vectors.
Trained Encoder / Decoder LSTM model with 15 epochs.
Predict the distractors using the concept uses in sequence generation ( at a
time predict only one word in loop and wait till the end token comes ).
Not as good quality of distractors formed due to lack of attention.
Fact Vs Opinion Statement Classification
Prepare the dataset and labeled it to fact vs opinion stating that the
sentences containing value words are mostly opinion.
Used Glove vectors to convert text to embeddings.
Trained LSTM using 5 epochs and binary cross entropy loss with 90% test set
accuracy.
Documents Clustering
Used K-means clusering to cluster the docs after preprocessing of text data.
Preprocessing includes tokenization of sentences and word2vec model to
get the feature vectors of words.
Captcha Solver
After preprocessing of captcha image used contour detection algorithm to
seperate digits.
Pass the seperated digit images one by one in trained deep learning model
on MNIST dataset.
News Headline Recommendation
After cleaning of headlines calculate aggregate similarity scores using word
overlap similarity and cosine similarity.
Sort the headlines in descending order of scores and use some threshold to
find relevant ones.
CERTIFICATES
Machine learning
Stanford University
Deep learning Specialization
deep learning.ai
Introduction to Data Science in Python
University of Michigan
Applied Machine learning in Python
University of Michigan
Applied Text Mining in Python
University of Michigan
Discipline
Responsibities
Achievements/Tasks