PERSONAL INFORMATION:

Name: Siddhartha Shrestha

Date of Birth: 12-11-1998
Address: Kalimati, Kathmandu
Mobile : +977-9808799274
Email: [email protected]
Linkedin profile: https://www.linkedin.com/in/siddharthashrestha
Github profile: https://github.com/sids07/

PROFILE
I am an enthusiastic and curious learner continuing my development in having a career in
Artificial Intelligence and Python. I believe the key to success is none other than hard work
and curiosity for learning and making new knowledge. I have a proactive attitude and
positive engagement with people.

EDUCATION

LEVEL INSTITUTE YEAR PERCENTAGE

Bachelor in Computer Kathmandu 2016 - 2020 70%

Engineering Engineering College

Higher Secondary Trinity International 2014-2016 76%

College

SLC Tarapunja English 2014 78%

Boarding School

EXPERIENCE
● Software Engineer, NLP:
Currently, I am working as a NLP Engineer on Treeleaf.ai. Here, i have been
working on several NLP projects and functionality such as chatbots , kgraph,
bot import, summarization, paraphrase, question generation, ticket suggestion
etc using different technologies such as python, java, arangodb, grpc, etc.
● Data Scientist at Wiseyak Inc.:
Here, I worked for 10 months including a 3-month internship. During my
period I got a chance to dive deep into the foundations of Machine Learning
such as probability, statistics, and linear algebra. Further, I researched various
Machine Learning Algorithms such as Linear Regression, Logistic Regression,
SVM, Decision Tree, Random forest, Adaboost, Neural Network, CNN, RNN,
etc with the mathematics behind it and its implementation.

Also, I got the chance to work with the data visualization team. Here I got a
chance to learn REACT JS and their visualization frameworks (like Victory JS,
React-Flow, vis.js) along with data manipulation and extraction for
visualization such as pandas, pdf plumber, etc. Further, using dash and its
framework VISDCC(which is using vis.js), I created a tree-structured graph
with nodes and edges along with expanding and collapse function.

● Jr. Python Developer:

During my 3 month period, I worked and learned on the following tools and
technology:
1. Werkzeug: To get knowledge on how web development actually works on
the lowest level.
2. Django (Web Development, REST framework): I learned to build
full-running websites using many Django features and APIs. Here also I got
familiar with database relationships such as one to one, one to many, many to
many, etc.
3. Javascript(SPA, API Fetch): Here I learned how the API created can be
accessed and shown on the frontend. Also, how can we make our application a
single page application we don't need to reload various HTML pages
frequently, we simply change the content of the single page?
4. Odoo: Here I got some familiarity with odoo framework how it works and
the various functions available on it.

● Computer Vision Nanodegree

During the first phase of lockdown, Udacity has offered free 1-month access to
all the course across Europe. So with the help of my friend in Europe, I got a
chance to get enrolled on Udacity Computer Vision Nanodegree where I got a
chance to learn from the basics to advanced concepts of Computer Vision
starting from CV2 to CNN to attention to localization. Further 3 projects were
 assigned to me as a part of a Nanodegree to compile the concepts I have
learned from this Nanodegree.

PROJECTS
● Chatbots:
Recently i have worked on chatbots with many functionalities. For intent
classification i have used rasa framework with DIET classifier. Further to add
functionality, i have prepared paraphrase models if user or client wish to add
more training examples. Another major functionality was bot import where we
give user flexibility and automation to build chatbot for their company by
simply providing company’s websites URL and our system automatically
scrape data from their site and using question generation, coreference resolver
and summarization model prepares the training phases, intents and bot
responses.

And adding multilinguality my another task and i did all of the mentioned
functionality on 4 languages i.e. english, spanish, nepali and hindi.

● AutomatedReply(Knowledge Graph):
Here, i got chance to learn three new technologies i.e. JAVA, ArangoDB and
Grpc in details. And using all 3 technologies we created options for user to
have fixedReply and resolutionReply via nodes and edges. My roles here we to
make sure all the system is working well as required, optimize the pipeline and
create the well rounded document of the workflow.

● Research on modern CNN architecture:

Here, I went through several modern CNN architectures such as VGG,
RESNET, DENSENET, NiN, etc, and researched what type of kernels, blocks
are being used and why it is used with its importance. Also, using 1*1, 1*3,
3*1 kernel and then concatenating to reduce the parameters of our models to
train faster with comparable accuracy. Finally, use of class activation
maps(CAM) and Gradient - CAM to visualize which portion of the image is
being focused on by our models.

● Symptom Similarity and Disease Prediction:

Here, I first preprocessed our Disease and Symptom data I took from
http://people.dbmi.columbia.edu/~friedma/Projects/DiseaseSymptomKB/index.html .
 then use Glove embedding along with the skip-gram model to train the data to get the
optimized embedding values and using cosine similarity we found similar symptoms.
And, also based on the symptoms given I used Multinominal Naive Bayes Algorithm
to detect the disease with respect to the probability.

● Image Captioning:
It was my second project on Computer Vision Nanodegree. Here seq2seeq model was
used i.e. encoder-decoder. In the encoder pre-trained CNN model resnet50 was used
and in the decoder, the LSTM model was used. Models were trained on GPU
provided on udacity workspace and on the COCO dataset which has 5 captions for
each image. After training loss of around 2.2 was obtained. And was able to get a
caption for the images.

● Facial Keypoint Detection:

During my enrollment in Computer Vision Nanodegree on udacity. It is my first
project for graduation. Here using facecascad e.xml from cv2 we first detected the
face with rectangular bounding boxes. Then using the CNN model we trained the
image data with 68 facial keypoints via the PyTorch framework applying all the
transformation to the input image as reshaping, random crop, tensor, normalize, etc .
And later, predict facial keypoints on test data. Further, we checked how the filter in
the convolution layer was used to extract features for the image.
Trained with avg loss: 0.05.

● SMS SpamClassifier:
Here, the classification of SMS to spam or ham was done. Firstly cleaning of data was
done by removing stopwords, punctuations, multiply white spaces, etc using a regular
expression. Then we tokenized the SMS data given in training data followed by
stemming and lemmatizing using NLTK. Further vectorizing of data was done as the
algorithm doesn’t take text data as input. For that various method was tried i.e. Count
Vectorization, N-grams, and Tf-Idf Vectorization. Moreover, to increase features we
did features engineering by placing two new features i.e. sentence length and number
of time punctuation used in raw text. And verified using matplotlib if it would be
useful to verify spam classifier then perform power transformation to add distribution
to that data. Finally using two ML algorithm gradient boosting and Random forest
respectively, we train the preprocessed data to classify SMS. And compare the result
which gives more accurate with accurate hypermeter using GridSearchCV.

● Transformer:
Here, Language translation was done from English to Spanish (as the Nepali dataset
was difficult to get). Seq2Seq model was used i.e. encoder-decoder but instead of
using RNN we just used a feed-forward network. And the main focus was the new
 technique Attention. And based on the paper “Attention is all you need” we prepared
the transformer model which was found to be quite faster as no RNN or heavy layer
was added.

SKILLS
1. Machine Learning
2. Deep Learning(Keras and Pytorch)
3. Computer Vision
4. Natural Language Processing
5. Probability and Statistics
6. Python
7. JAVA
8. ArangoDB
9. Grpc
10. Django
11. Basic React.JS (Victory.JS, ReactFlow)
12. Basic JS
13. Bootstrap