Pratyusha Parashar

Pratyusha Parashar

About

Detail

Data Scientist
United States
Contact Pratyusha regarding: 
work
Full-time jobs
Starting at USD100k/year

Timeline

work
Job
school
Education

Résumé

Jobs verified_user 0% verified
  • Northeastern University
    Researcher
    Northeastern University
    Nov 2023 - Apr 2024 (6 months)
    • Deployed vectorized data pipelines leveraging PostgreSQL, Microsoft Azure, and FAISS for efficient data ingestion, similarity searches, and retrieval, enabling data-driven decision-making with 43% speedup.
    • Leveraged techniques like Low-Rank Adaptation (LORA) and Supervised Fine Tuning (SFT) to finetune Llama2 and achieved 45% improvement over the pretrained model for downstream NER and relation extraction.
  • Castleton Commodities International
    Data Science Intern
    Castleton Commodities International
    Jun 2023 - Aug 2023 (3 months)
    • Developedand deployed a scalable content extraction platform using LLMs onAWS to streamline information retrieval process for trading desks, aimed at structuring and summarizing data to generate new datasets for further analyses from diverse document formats sourced from multiple channels in several languages saving analysts countless hours of manual information extraction.
    • Utilized and evaluated RAG to build a multi-document Question Answering system leveraging Weviate Vector Database to implement use-cases for Oil, Power and Natural Gas trading desks, HR, and IT teams.
    • Implemented OpenAI's function calling feature to extract structured data from documents using static and dynamic prompting techniques.
    • Conducte
  • Vertex Pharmaceuticals
    Data Science Co-op
    Vertex Pharmaceuticals
    Jan 2023 - Jun 2023 (6 months)
    • Led development of a ML system using an ensemble of Logistic Regression and Random Forests, predicting job requisition difficulty, enabling optimized allocation of recruiter workload & enhanced performance tracking capabilities, improving RMSE by 40%.
    • Leveraged GPT-3.5 to engineer a robust joint entity and relation extraction system with 98% accuracy, specifically tailored for the executive search team's requirements in identifying prospective candidates for research roles.
    • Designed, developed and deployed an advanced real-time Power BI dashboard to visualize and analyze pay ranges within job postings in pharma, offering a holistic view of the labor market and enabling in-depth analysis of competitor salaries and indust
  • Vedantu
    Product Associate
    Vedantu
    Aug 2021 - Jan 2022 (6 months)
    - Monitored and analyzed product performance metrics through A/B tests to compare variants, as well as investigating anomalies across Automation, Customer Experience (CX), and Child Safety domains, delivering actionable insights for product enhancement.
    - Partnered with interdisciplinary teams spanning analytics, product management, and data science to meticulously document, validate, and scrutinize data integrity, while conducting thorough testing protocols to guarantee seamless integration of new functionalities.
    - Led initiatives to enhance the efficacy of customer service chatbot, resulting in a notable 30% upsurge in both query submission and response volume.
    - Implemented proof of concept for a profanity filter in c
  • R
    Data Science Intern
    Reserve Bank of India (India's Central Bank)
    Jan 2021 - Mar 2021 (3 months)
    • Finetuned a transformer model (FinBERT) using LAMB optimizer for news sentiment analysis to predict public opinion in real time on the economic variables, 10 Indian Banks and Non-Banking Financial Companies, and achieved an F1-score of 87%, all of which had/is being used as an input in formulation of monetary policy.
  • M
    Machine Learning Engineer Intern
    Mavoix Solutions
    Aug 2020 - Nov 2020 (4 months)
    • Developed an image classification model with a MCC of 89% on a custom dataset for identifying skin infection severity, employing techniques such as classic + mixup data augmentation, cosine annealing learning rate schedules, and label smoothing.
  • Vidyalankar Institute of Technology
    SoftwareDeveloperIntern
    Vidyalankar Institute of Technology
    Jul 2020 - Aug 2020 (2 months)
    • Developed and deployed a real-time facial recognition model using AWS Lambda, achieving a remarkable 98% F1 score, facilitating a contactless attendance management system serving over 8000 users.
  • K
    Machine Learning Engineer Intern
    Kreatio Software
    Jan 2020 - Jun 2020 (6 months)
    - Built a recommender system to suggest different visualization tools based on the VizML paper and incorporated deep learning models in the company’s analytics product (NJAA)
Education verified_user 0% verified
  • Northeastern University
    M.S. in
    Northeastern University
    Jan 2022 - Dec 2023 (2 years)
  • University of Mumbai
    B.E. in
    University of Mumbai
    Jun 2017 - Jul 2021 (4 years 2 months)
  • D
    Data Engineering Zoomcamp, NLP Specialization
Projects (professional or personal) verified_user 0% verified
  • Q
    QuantizedCNN
    Pruned (over weights) + QuantizedVGG-16 32-bit model to 2-bits (32x reduction in model size) using Fine grained Pruning and K-Means Quantization while retaining 98% accuracy through subsequent finetuning. Forecasted Rossman store sales and achieved top 1% on the private leaderboard using ensemble of XGBoost and a tabular deep learning model. Developed supervised models that reliably estimate prediction uncertainty using AUSE and ECE as KPIs.