Soham Sarfare

Experience

• 

  Indian Institute of Management Ahmedabad

  Feb 2017 - Apr 2017

  Research Assistant

  * Worked as a Research Assistant to the RTERC cell of IIM Ahmedabad.* Gathered data on the implementation of the RTE section 12-1-c in the state of Maharashtra, India. * Verified data by interviewing government officials. * Organised and visualised data which was subsequently published in "State of the Nation RTE Section 12(1)(c)" - 2017

• 

  Fiverr

  Apr 2018 - Dec 2019

  Freelance Web Developer

  Designed and develop Websites to specifications provided on the freelance platform Fiverr.https://www.fiverr.com/sohamsarfare

• 

  Truuth

  Feb 2021 - now

  • Built a data transformation pipeline to extract and parse tabular data from invoices. Utilized Computer Vision models to OCR data which is then parsed using name entity natural language processing models. This approach has achieved a straight through output of 70% leading to reduced manual review.• Train, test and validate a fusion classification deep learning model to accurately classify identity documents. The model utilizes scikit-learn and pytorch frameworks in python. A final classification accuracy of 96% was achieved on Australian identity documents. Deployed the model for inference as a containerized docker Flask application on AWS ECS + EC2.• Conduct literature review and identify approaches to detect tampering in identity documents. Developed a deep learning model utilizing pytorch framework with Sagemaker instances, achieving an accuracy of 95% on detecting tampering on Australian identity documents. Deployed the model for inference as a containerized docker Flask application on AWS ECS + EC2.• Constructed an algorithm to identify repeat applicants and analyze the possibility of fraud. The algorithm utilizes facial recognition combined with ocr techniques to extract and analyze applicant data. On review of historic client data, this algorithm was able to identify previously unknown fraudulent actors.• Developing CI/CD pipelines for machine learning applications. Utilizing AWS cloudstack with serverlessand cloudformation templates with Travis as executor. Show less

  • Data Science Consultant

    Jun 2021 - now

  • Data Science Intern

    Feb 2021 - Jun 2021