Ashutosh Kumar Email & Phone Number

California, United States

Thousand Oaks, Ca, Us

Irvine, Ca, Us

• Developed DAADT, a groundbreaking Disease-Associated Antibody Discovery Tool, delivering end-to-end solutions with publishable results.• Engineered scalable REST APIs in Python Flask for diverse scenarios within DAADT, fostering flexibility, efficiency, and facilitating seamless communication between front-end, back-end and database components through collaborative discussions with cross-functional Researchers.• Built the User Interface independently using ReactJS, creating a seamless and user-friendly experience for Stakeholders and Researchers.• Designed and implemented a 218GB human, bacteria and virus protein database in AWS RDS using bash scripts, optimizing data ingestion through ETL pipelines, and transforming raw text sequences into a MYSQL database.• Configured the Database connector in Flask to ensure robust data flow, resulting in a 50% enhancement in accessibility compared to Fasta files and a 30% reduction in data loss.• Leveraged cloud technologies to deploy the entire application as containers on AWS EC2 instances using Docker, optimizing resource utilization and scalability for enhanced performance.• Pioneered APDT (Alzheimer Plaque Detection Tool) to detect and quantify Alzheimer's plaques from brain scan images using OpenCV and PIL. Achieved a substantial 90% accuracy boost in comparing before and after treatment features, significantly enhancing research efficiency.

Pune, Maharashtra, In

• Created a comprehensive Early Breast Cancer classification solution, covering Ductal Carcinoma in Situ (DCIS) detection.• Engineered RESTful APIs for efficient model training pipeline, integrating seamlessly with the ReactJS user interface for data analysis presentation.• Utilized FastAPI and Flask to create a Whole Slide Image (WSI) processing and segmentation pipeline, handling WSI sizes from 600MB to 1GB.• Architected, built, connected, and deployed a deep learning DCIS classifier, leveraging machine learning (XGB) and OpenCV for feature extraction and implemented Transfer learning using ResNet-50 and Densenet-131 with PyTorch for image classification training.• Deployed the entire system using GCP Compute Engine and Docker, ensuring scalability and efficient utilization of cloud resources.• Achieved an 86% accuracy rate in DCIS classification, with a recall score of 0.88, 0.77 precision, and an F1 score of 0.82, leading to the success of the project and serving as a Proof of Concept for numerous pathology-related AI-enabled projects, contributing significantly to the company's growth in the health sector and medical AI.

La Jolla, California, Us

• Devised a Dynamic NLP-Powered News Visualization System, constructing a robust ETL pipeline with Apache Spark and Pandas to handle over 100k news articles. • Developed an NLP pipeline utilizing Spacy, NLTK, Flair, and PyTorch for comprehensive text preprocessing, cleaning, lemmatization, and coreference resolution. Conducted Named Entity Recognition (NER), Parts of Speech Tags (POS) and Relation extractions. Implementing advanced topic modeling using Gensim and information retrieval techniques, leveraging semantic word vectors from Word2Vec.• Allowing users to seamlessly search and visualize news on a world map while exploring the intricate network between events and subevents.

● Engineered a Live Neural Chatbot integrating Google Dialogflow, Flask, NLTK, and Stanfordcorenlp for entity linking. Enabled factual question answering by generating SPARQL queries from neural machine-translated questions, accessing the DBpedia Database. Also, revamped 80% of the Neural SPARQL Machines repository, enhancing TensorFlow-based NMT (>2.0) with improved architecture and performance, attracting 177 stars and 85 forks● The chatbot can answer factual questions by extracting entities and relations from the question, forming a SPARQL query, and fetching answers using the query from DBpedia Database.

Chandigarh, Chandigarh, In

• Launched an NLP-powered tool for cold emailing, leveraging generative AI to achieve a remarkable 40% increase in email opening rates.• Orchestrated the Sequence-to-Sequence Generation with Attention, fine-tuned LLMs (BERT and BART) for Question Answering, Prompting GPT for summarization, and employed advanced techniques such as Entity Extraction using NLTK and Spacy, and Coreference Resolution using AllenNLP. The integration of Hugging Face, PyTorch and CUDA for implementation, transforming the email outreach process and enhancing user engagement.

Master'S Degree, Computer Science

Bachelor Of Technology - Btech, Computer Science

High School, Science