karan

karan

+•• ••••• •••••••••••@•••••.com

3.5 yrs

As a data scientist with 6 years of experience, I am driven by a deep ambition to master complex analytical challenges and deliver precise, impactful solutions. I thrive in collaborative environments, leveraging my resourcefulness and flexible approach to adapt to evolving data landscapes and continuously refine my craft.

  • Open to Mumbai · Pune · Bangalore
  • On-site, Hybrid, Remote

Ask about karan

Skills

Claimed

TextBlob UsageData ScienceSelf-Directed LearningProblem SolvingAdaptabilityWork Planning and PrioritizationTransparent CommunicationTranslationNatural Language GenerationGPT-2 UsageTraining Data CreationHandling Data Imbalance

Proof of Experience

Fractal Analytics

Data Scientist

Fractal Analytics · Sep 2022 – Mar 2023

  • Built end-to-end sales forecasting pipeline and achieved 87% accuracy for the retail domain, managing data storage in MySQL, training an OLS regression model on two years of time series data with 17 engineered features
  • Engineered predictive features including time-trend variables, calculated price and discount elasticities using OLS regression coefficients to quantify the impact of pricing decisions on sales performance
  • Collaborated cross-functionally with the business team to develop a pricing scenario simulation framework, to improve promotional ROI by 12% while reducing unnecessary discount spend by 8%
  • Translated complex model outputs into actionable business recommendations, empowering pricing and trade marketing teams to make insight-driven decisions on discount depth and promotional planning

Data Scientist

SPi Technologies India Pvt Ltd · Aug 2021 – Sep 2022

  • Trained 13 spaCy v3 NER models to extract ESG-related entities from banking-domain PDF reports, achieving an average F1-score of 88% and reducing manual data extraction effort by 62%
  • Developed a PyTorch-based Page Classification model achieving 92% recall across 1,500 PDF documents to filter relevant pages prior to NER processing, reducing end-to-end processing latency by 34% and lowering computational costs by 27%
  • Enhanced entity extraction accuracy by training a YOLO object detection model to identify high-density table regions and deploying a DistilBERT-based Question Answering model for complex entity extraction, improving overall entity capture rate by 15% and increasing precision for critical KPIs
  • Deployed all NLP models on AWS using Celery for asynchronous task management, enabling scalable batch processing and reducing document turnaround time to 10 minutes per document
  • Built an end-to-end YOLO object detection pipeline to detect 20 object types from public space imagery, achieving 87% detection accuracy and enabling automated risk assessment workflows that reduced manual inspection effort by 40%
Straive

Data Scientist

Straive · Aug 2021 – Sep 2022

  • The company was formerly known as SPi Global/SPi Technologies India Pvt Ltd.
  • 1.
  • Trained 13 spaCy v3 NER models to extract ESG-related entities from banking-domain PDF reports, achieving an average F1-score of 88% and reducing manual data extraction effort by 62%
  • 2.
  • Developed a PyTorch-based Page Classification model achieving 92% recall across 1,500 PDF documents to filter relevant pages prior to NER processing, reducing end-to-end processing latency by 34% and lowering computational costs by 27%
  • 3.
  • Enhanced entity extraction accuracy by training a YOLO object detection model to identify high-density table regions and deploying a DistilBERT-based Question Answering model for complex entity extraction, improving overall entity capture rate by 15% and increasing precision for critical KPIs
  • 4.
  • Deployed all NLP models on AWS using Celery for asynchronous task management, enabling scalable batch processing and reducing document turnaround time to 10 minutes per document
Shyena Tech Yarns

Data Scientist

Shyena Tech Yarns · Sep 2019 – Aug 2021

  • Started here as an intern and was subsequently offered a full time position
  • Developed a multi-level text classification system for Dutch municipal complaint data, progressing from TF-IDF + SVM/Logistic Regression baselines to a BERT-based Deep Neural Network, achieving 84% accuracy and improving complaint routing efficiency by 42%
  • Deployed the classification model on AWS EC2 using Docker, enabling a scalable, containerized environment for real-time complaint processing
  • Built a semantic product-matching system using BioBERT embeddings and cosine similarity, improving product search relevance by 27% and reducing manual catalogue matching efforts by 31%, enhancing operational efficiency for inventory management teams
  • Developed and deployed a YOLO-based object detection model on AWS EC2 for real-time estimation of age, gender, and emotions from live video feeds, achieving 87% detection accuracy and enabling store managers to analyse customer demographics and engagement patterns
  • Translated model-generated demographic insights into actionable customer segmentation strategies, driving an estimated 15% improvement in campaign effectiveness for retail marketing teams

Proof-of-Work

No completed sessions yet. Anything karan finishes shows up here with the evidence behind it.

Proof of Education

Masters, Applied Data Science and Analytics

SRH Hochschule Heidelberg, Germany · 2023 — 2025

Bachelor, Electrical Engineering

University of Pune, India · 2015 — 2019

A Proof CV — one profile, kept current, with the proof attached. Back to the portal