Live Stream

Candidate Details

****** ****
Strong Hire
**********@gmail.com
+91-**********
27-04-2023
6:00 PM
Data Scientist (NLP) Ujjwal
4.36
ADA
23% Behaviour Skill Set
77% Technical Skill Set Mandatory

Behaviour
Skill set

Communication
3.0
Industry Awareness
5.0
Engineering Mindset
5.0
Attitude
5.0
Team Work
5.0
Problem Solving
5.0
Analytical Skill
5.0

Technical Skill Set
Mandatory

C++
5.0
MongoDB
5.0
Machine Learning
5.0
NLP
5.0
MySQL
5.0
GCP
2.0
Bash
3.0
Pytorch
5.0
Keras
5.0
Pandas
5.0
Numpy
5.0
Tensorflow
5.0
BERT
5.0
Precision
5.0
Recall
5.0
WER
5.0
BLEU Score
3.0
ROUGE
3.0
METEOR
3.0
AUC-ROC
5.0
F1-Score
5.0
Coherence
5.0
Plotly
5.0
Matplotlib
5.0

The candidate is very good and has a strong foundation in the fundamentals of NLP. Although the candidate was unable to answer some of the theoretical questions and also needs improvement in basic communication I still feel that he will be a perfect fit for the role as per the JD given. There could be a learning curve associated with his hiring, but he will be able to pick it up soon and become an independent contributor to the team. I still feel that given his previous experience, he should be the right resource for this job.

Excellent

Good

Satisfactory

Needs Improvements

Unsatisfactory

Communication

Satisfactory

3.0

the candidate is a little slow in communication but this should not be a deal breaker

Industry Awareness

Excellent

5.0

the candidate has very good knowledge of the industry and the leading techniques used

Engineering Mindset

Excellent

5.0

the candidate has a very good mindset for tackling engineering problems

Attitude

Excellent

5.0

the candidate was able to demonstrate a good attitude during the interview process

Team Work

Excellent

5.0

the candidate has worked on multiple projects in teams before and has the relevant experience

Problem Solving

Excellent

5.0

during the coding round, the candidate was able to very quickly understand the given problem and solve it

Analytical Skill

Excellent

5.0

the candidate was able to showcase very good analytical thinking while discussing various NLP topics

Excellent

Good

Satisfactory

Needs Improvements

Unsatisfactory

Technical Skill Set Mandatory

C++

5.0

The candidate has multiple years on experience in C++

MongoDB

5.0

The candidate has previously worked with mongoDB in multiple projects

Machine Learning

5.0

The candidate has previously worked with ML techniques in multiple projects

NLP

5.0

The candidate's core strength is NLP as he has worked on many projects in the past.

MySQL

5.0

The candidate was able to convey that he as the relevant experience of working on MySql

GCP

2.0

The candidate has no prior experience working in GCP

BASH

3.0

The candidate has very limited experience working in bash

Pytorch

5.0

The candidate has worked on many projects that involved working on Pytorch

Keras

5.0

The candidate has worked on many projects that involved working on KERAS

Pandas

5.0

The candidate has worked on many projects that involved working on PANDAS

Numpy

5.0

The candidate has worked on many projects that involved working on NUMPY

Tensorflow

5.0

The candidate has worked on many projects that involved working on TENSORFLOW

BERT

5.0

The candidate has worked on many projects that involved working on BERT

Precision

5.0

The candidate is aware on how precision works in ML optimization techniques

Recall

5.0

The candidate is aware of how recall works in ML optimization techniques

WER

5.0

The candidate has previously worked on WER and has worked on many similar projects.

BLEU Score

3.0

The candidate has very limited experience in this field

ROUGE

3.0

The candidate has very limited experience in this field

METEOR

3.0

The candidate has very limited experience in this field

AUC-ROC

5.0

The candidate is aware of how AUC-ROC works in ML optimization techniques

F1-Score

5.0

The candidate is aware of how F1 score works in ML optimization techniques

Coherence

5.0

The candidate is aware of how the coherence technique works in ML optimization

Plotly

5.0

The candidate has worked on many projects that involved working on plotly

Matplotlib

5.0

The candidate has worked on many projects that involved working on matplotlib

Q-

Output script

hello, world a2b3c1 a3b3 Traceback (most recent call last): File "script.py", line 34, in import numpy as np ModuleNotFoundError: No module named 'numpy'

Question 1

How are Word2Vec embeddings different than GloVe embeddings?

Candidate Answer

Word2vec embeddings are based on training a shallow feedforward neural network while glove embeddings are learnt based on matrix factorization techniques

Question 2

What is the difference between having an auto-encoder or auto-regressive models?

Candidate Answer

The only difference is how the model is pre-trained. The same architecture can be used for both the models.

Question 3

What are the advantages of using BiLSTM over LSTMs?

Candidate Answer

In Bidirectional LSTM we can input the data in both the direction and hence it is more powerful than LSTM.

Question 4

Was IR used in knowledge graph, what algorithms were used?

Candidate Answer

A knowledge graph is a representation of entities (e.g., a person, a place) and the relationships between them. Knowledge graphs are used to derive semantic understanding from these connections.

Question 5

What prompting techniques were used in LLMs?

Candidate Answer

Prompts are the inputs or queries that a user or a program gives to an LLM AI, in order to get a specific response from the model. Prompts can be NLsentences or questions, code snippets/commands, combination of text or code, depending on the domain and the task.

Question 6

Naive Bayes implementation?

Candidate Answer

It is used in sentiment analysis

Question 7

What are counterfactual generative models?

Candidate Answer

Couldn't clearly tell about the model, ended up explaining conditional generative model

Question 8

Basic architecture of CNNs

Candidate Answer

It has three layers namely, convolutional, pooling, and a fully connected layer. CNN have grid like typology.

Question 9

What are the differences between RNNs, GRUs and LSTMs

Candidate Answer

the key difference between RNNs, LSTMs, and GRUs is the way that they handle memory and dependencies between time steps.rnRNNs remember information from previous inputs but will become very slow with long-term dependencies. LSTMs can store and deal with long-term dependencies using a special type of memory cell and gates. GRUs is simplified version of LSTMs and uses single update gate and is therefore easier to train and run.

Question 10

What are the differences between tokenizers from nltk and huggingface tokenizers?

Candidate Answer

Candidate has little information on the difference between them. He was able to explain how tokenisation works.

Question 11

GRU and LSTMs difference in architecture

Candidate Answer

[similar question]Already discussed above.

Question 12

GRU and LSTMs difference in architecture

Candidate Answer

[same question as above] Already discussed above.

Question 13

Transformers vs RNNs

Candidate Answer

Transformers also make use of attention mechanisms but RNN does not.

Question 14

What is self attention?

Candidate Answer

In a single sentence, how the words are related to each other. this is gathered by Self attention.

Question 15

Different ways to get word embeddings?

Candidate Answer

One-hot encoding, tf-idf, word2vec, fasttext

Question 16

Do vectors hold semantic meanings?

Candidate Answer

Candidate was able to answer YES. But could not go in detail when asked to take an example and explain.

Question 17

LLMs are said to have a lot of parameters, what does it mean?

Candidate Answer

the parameters are basically the variables present in the ML to be trained.

Question 18

What are various optimization algorithms to decide the learning rate?

Candidate Answer

adam optimisation, learning rate decay, gradient descent

Question 19

How is BERT different than GPT-3?

Candidate Answer

gpt-3 is an autoregressive model, while BERT is bidirectional

screenshot 1

Overall Summary

The meeting started with Candidate introducing himself as a front-end developer with expertise in React and JavaScript. He highlighted his experience in backend development and team leadership. The discussion focused on technical skills, React, lifecycle hooks, state management with Redux, and ES6. Candidate also provided insights on API service calls, unit testing, and security practices. The meeting ended with action items assigned to Candidate for further discussion and assessment.

Additional Data Points

  • Coordinator
  • ADA
  • Candidate
  • Introduction
  • Technical Skills
  • React
  • Lifecycle Hooks
  • Functional Component vs Class Component
  • Pure Component
  • API Service Call
  • State Management with Redux
  • ES6
  • Babel
  • Webpack
  • Map, ForEach, and Filter
  • Closure
  • Event Bubbling and Event Capturing
  • Javascript Prototype
  • Variable Hoisting and Method Hoisting
  • Debouncing and Throttling
  • Cross-Origin Resource Sharing (CORS)
  • API Security
  • HTML5 Features
  • Securing the API for unauthorized access
  • Latest features used from HTML5
  • Difference between block element and inline element
  • Difference between cookies, local storage, and session storage
  • Pseudo classes in CSS
  • How to develop a responsive web design
  • Backend technologies used
  • Experience with graphic design tools
  • Problem-solving exercise
  • Improving application performance
  • Key features of Python
  • Advantages of Figma as a designing tool
  • Prototyping in Figma
  • Candidate introduced himself as a front-end developer with 9 years of experience
  • Candidate has expertise in React, JavaScript, CSS, Material UI, and Angular
  • Candidate also has a certificate in AWS Associate Developer
  • Candidate has experience in backend development and team leadership
  • Candidate highlighted his strengths in pixel-perfect responsive UI and performance optimization
  • Discussion on React versions used and working with functional components
  • Difference between single-page and multi-page applications
  • Explanation of virtual DOM and real DOM in React
  • Overview of React lifecycle hooks in class components
  • Comparison between functional components and class components
  • Utilization of pure components for performance optimization
  • Methods for making synchronous API service calls using Axios and async/await
  • Overview of state management libraries like Redux Saga, Redux Thunk, and Zustand
  • Explanation of Redux and how it works with reducers and actions
  • Difference between controlled and uncontrolled components
  • Discussion on writing unit test cases for login component
  • Dependency management in package.json: dev dependencies vs dependencies
  • Advantages of ES6 in JavaScript
  • Explanation of Babel and its purpose to convert ES6 code to browser-understandable version
  • Webpack's role in bundling JavaScript, HTML, and CSS files
  • Difference between map, forEach, and filter array methods
  • Usage of closure in JavaScript for storing variables
  • Explanation of event bubbling and event capturing
  • Utilization of JavaScript prototypes for object inheritance
  • Difference between variable hoisting and method hoisting
  • Explanation of debouncing and throttling in JavaScript for optimizing event-based operations
  • Overview of Cross-Origin Resource Sharing (CORS) and its use in API access control
  • Methods to secure APIs from unauthorized access
  • Discussion on the latest features of HTML5, including picture, header, and footer tags
  • To secure the API for unauthorized access, HTTPS can be used instead of normal HTTP
  • The latest features used from HTML5 include the picture tag, header tag, footer tag, and local storage and session storage
  • The doctype is used to determine the version of HTML being used and is mandatory
  • Block elements and inline elements behave differently in terms of line breaks and positioning
  • Cookies are sent with API calls, while local storage and session storage are not
  • Pseudo classes in CSS include active, focus, and visible
  • Responsive web design can be achieved using media queries and different CSS classes for different screen sizes
  • Backend technologies mentioned include Node.js and Python
  • Figma is the preferred graphic design tool, known for its ease of use and CSS property generation
  • A problem-solving exercise was conducted on finding repeated characters in a string
  • Various techniques were discussed to improve application performance, including CDN usage, hooks, image optimization, and preloading resources
  • Key features of Python were not discussed in detail
  • Advantages of Figma include ease of design, CSS property generation, and its acquisition by Adobe
  • The use of prototyping in Figma was not discussed in detail
  • Candidate to share the coding ID link in the chat box for coding assessment
  • ADA to assess Candidate's technical skills and project experience
  • Candidate to provide more details about unit test cases and API security practices
  • Coordinator to follow up with Candidate regarding the next steps
  • Investigate HTTPS for securing the API
  • Explore the latest features in HTML5
  • Research the differences between block and inline elements
  • Read more about cookies, local storage, and session storage
  • Learn about pseudo classes in CSS
  • Study media queries for responsive web design
  • Further explore backend technologies such as Node.js and Python
  • Improve knowledge and experience with graphic design tools
  • Practice problem-solving exercises
  • Implement techniques to improve application performance
  • Gain more understanding of key features in Python
  • Continue using Figma for graphic design projects
  • Learn more about prototyping in Figma