Notes

KULDEEP SINGH ARYA
[email protected]

Profile Summary

• Highly skilled Data Science Professional with 5.3 years of expertise in AI, ML, DL, NLP coupled with 1.11 years of Specialized Proficiency in Gen AI.
• Worked on many ML projects end to end on supervised and unsupervised learning.
• Worked on NLP project and solved several problems like text classification, sentiment analysis, summaries, question and answering using Nltk, Spacy, NER, Hugging Face, transformers models BERT etc.
• Worked on Generative AI application using different LLMs like Vertex AI, OpenAI, AWS Bedrock, RAG framework.
• Worked on to fine-tuning LLM models using Lora (PEFT) and Qlora techniques
• Working experience & extensive knowledge in python with libraries such as Numpy, Pandas, Matplotlib, Seaborn, Sklearn, Keras, Tensorflow, PyTorch, SpaCy, Nltk, OpenCv, Transformers, PySpark & LangChain.
• Have extensive knowledge of Scrum and Agile Methodology in Software Development.
• Experienced in end-to-end delivery of project from requirements gathering, design and implementation, and documentations.
• Excellent time management skills with ability to perform under pressure & meet deadlines.

Education

Bachelor of Engineering – Amity University - Haryana, India - 2017

Work Experience

Persistent Systems Limited - Lead Software Engineer
Noida, India – Mar 2024 to till date.

EAB
Duration: 3 Months
Technologies/ Languages used: Python, GenAI, OpenAI, Prompt Engineering, Networkx library, AWS lambda service, AWS Workspaces

Developed a Script for fetching complex SQL query response in a Json format using OpenAI model gpt4o, gpt4o-mini etc. find actual table detail in node and edges form and generate csv using script. After that add more static edges using networkx library. As well as extracting knowledge graphs with different school datasets.

Mission Cloud Chatbot_POC
Duration: 2 Months
Technologies/Languages used: Python, NLP, Lang chain, GenAI, Prompt Engineering, Amazon Bedrock, AWS Lambda Service, AWS EC2, Image Textract Services, Streamlit.

Developed a proof-of-concept Chatbot utilizes a Streamlit user interface. Users can upload a PDF document, and the application converts each page into a base64 encoded image. These images are then fed into Amazon's Bedrock Claude3 model, a powerful Large Language Model (LLM). The user can then ask questions about the document, and the Chatbot leverages the user's specific prompts in conjunction with Claude3 to extract relevant information and provide answers.


Chetu India Pvt. Ltd. - Software Engineer
Noida, India - Oct 21 to Mar 24


Public Private AI Chat Agent for Sales CRM
Duration: 1.6 years.
Technologies/Languages used: Python, NLP, Deep Learning, Lang chain, GEN AI, RAG, OPENAI, Azure, LLM, Vector DB, Prompt Engineering.

Developing an advanced public private AI chat agent tailored for Sales CRM company knowledge source, our project integrates NLP, deep learning, OpenAI, Lang chain, Vector DB like Chroma, Faiss, Weaviate, and pinecone, also used LLM models. This agent adeptly comprehends client queries, offers accurate information, and provides support to clients. Our model is trained to sales crm product and services related information and generate contextually relevant responses, ensuring accurate and meaningful interactions. For Evaluation we are use Bleu Score, ROUGE Score, ROUGE-N, DeepEval. By harnessing deep learning, the agent interprets complex queries, extracts vital information, and responds knowledgeably, streamlining user support and communication. This Public Private AI chat agent enhances user and client engagement, streamlines communication, and empowers institutions to offer efficient support and sales information dissemination. Our project underscores AI's transformative potential in sales CRM, merging NLP, deep learning, and medical proficiency for enhanced user interactions and improved client outcomes.


POC: Blog Generation using LLM Model
Technologies/Languages used: Python, ML, DL, NLP, AWS Bedrock, LLAM2 70B, Prompt Engineering like zero shot.

Evaluate the feasibility of automating blog post creation with LLMs. Explore the potential of AWS Bedrock for integrating various LLM functionalities into content creation workflows. Demonstrate the capability of generating different blog post formats and styles using LLM prompts. Successfully generated blog posts on diverse topics using LLM prompts, showcasing the model's ability to handle a variety of content.
Achieved human-quality writing in terms of [mention specific aspects - e.g., grammar, coherence, factual accuracy] for a significant portion of generated content.
Identified potential for increased content creation efficiency and reduced production time.
LLMs for automating content creation tasks such as blog post generation. For this content generation we use Context Precision, Context Relevancy and faithfulness evaluation metrics. AWS Bedrock as a platform for integrating advanced NLP functionalities into content workflows and fine-tuning using (PEFT) Lora techniques for llama2 model. Customizing LLM outputs through effective prompt design to achieve different content styles and formats.


Image Classification
Duration: 4 Months.
Technologies/Languages used: Python, ML, DL, AWS Sege maker, AWS S3.

This project focuses on deploying an image classification system for Tyre conditions in three classes Good, Average and Bad. Leveraging deep learning models CNN Model, VGG16. for image storage using S3 Bucket and deployed on AWS Sege Maker make an API endpoint for cross platform like mobile application. The incorporation of behavioral analysis ensures the detection of irregularities, minimizing the need for manual intervention.


Multivariate Analysis for a cell analysis
Duration: 6
Technologies/Languages used: Python, ML, Flask, statistics, AWS.

In this project, we developed a predictive model using csv data. Leveraging ML models and using statistical analysis performed on numerical data. Calculate how much virus effected on cell in human blood using PCA analysis. Deployed on multiple flask API on AWS EC2 instance for creating multivariate analysis report.





Fourtek IT Solutions Pvt. Ltd. - Software Engineer- ML
Noida, India - June 19 to Sep 21


Domain wise Language Translation.
Duration: 2 Years
Technologies/Languages used: Python, DL, RNN, LSTM, BERT, Transformer Architecture, Transfer learning, Attention Mechanism, AWS EC2.

In this project we developed a language translation system using multiple Deep learning models with help of NLP Techniques. Collect translate parallel corpus like all Indian local languages English-Hindi, English – Marathi, English – Bengali, English- Tamil etc. We created and trained multiple language translation model using Transformer model and deployed on AWS EC2.
In the Language Translation industry, we provide multiple translation and transliteration services.

POC: Resume Parser NER with Classification based Skills
Technologies/Languages used: Python, ML, DL, NLP, AWS.

Extract key entities (e.g., people, organizations, locations) from text data. Demonstrate the feasibility of using NER for Resume. Achieved an accuracy of 80% in entity recognition. Successfully identified and classified various entity types. Identified potential applications of NER technology within the company for tasks Resume Selection. Using this System, we optimized HR Time and Cost and Create an automatic system for helping candidate selections process.











Key Skills and Knowledge
Python
Machine Learning
Deep Learning
Natural Language Processing
Gen AI
Prompt Engineering
Vector DB- ChromaDB, Pinecone, Faiss-CPU
AWS Bedrock, SageMaker, EC2, S3
Azure OpenAI


Other Skills
SQL


Tools
GitHub
Anaconda
PyCharm
Visual Studio