Utsav Mehta

Experience

• 

  Hacktoberfest

  Oct 2021 - Oct 2021

  Contributor
• 

  Meiro Mobility

  Oct 2021 - Dec 2021

  Software Development Intern

  Worked on developing mobile applications for this product-based company. One of the applications served the purpose of booking rides for customers and the other application was meant to be used by auto-rickshaw drivers.I mainly Designed a user landing screen, implemented foreground tasks for location recording in Android, and made an analysis interface for the auto-rickshaw drivers. Gained Experience and followed standard practices of Agile development and using product management kits like Trello. Show less

• 

  ISRO - Indian Space Research Organization

  Jun 2022 - Sept 2022

  Research Intern

  • Contributed to ”Navigation Signal Anomaly Detection” research, analyzed trends in time-series data with Plotly, & developed models like Random Forest, LSTM using scikit-learn & TensorFlow with 98% & 97% accuracy, respectively.• Engineered and deployed a real-time L5-signal classification model using FastAPI framework, implementing internal network accessibility through ngrok tunneling, which enhanced real-time positioning accuracy by 23% for the navigation system.

• 

  Differently Enabled Energy for Disabled (DEED) Mobility Solutions

  Jul 2022 - Aug 2022

  Software Development Intern

  • Developed an Android app for shared mobility in Java, utilized NFC for kiosk pairing and Firestore Realtime Database for dynamic ETA updates. Integrated Stripe for secure payments, significantly enhancing the user experience.• Engineered fare estimation with Firebase Functions using captured location, gyroscope data, and timestamps. Designed a dashboard in the driver’s app to visualize income and peak-hour stats using D3.js, boosting their weekly revenue by 15%.

• 

  Indian Institute of Technology Gandhinagar

  May 2023 - Jul 2023

  Research Fellow

  • Devised a Python library with PyTorch, SciPy, & statistical methods for active learning in regression tasks, reducing training time by 29% and data needs by 54%, cutting data-labeling costs, & saving significant time.• Developed models for mobile spirometer hardware to predict lung health from noisy respiratory audio. Used SciPy & NumPy for preprocessing, & implemented Wav2Vec transfer learning with PyTorch, along with 1D-CNNs and spectrogram analysis using TensorFlow, achieving an 87% accuracy in predicting lung health parameters Show less

• 

  TorchIt

  Dec 2023 - Mar 2024

  Machine Learning Intern

  • Led a team of 9 to develop an Android app for the visually impaired, integrating Quantized YOLOv8 for object detection, FaceNet model with SQLite to save embeddings for face recognition, & TFLite for color and currency detection.• Optimized scene description and interaction with a serverless function of vision transformer on AWS Lambda, with an optimal processing time of 850 ms achieved via image compression using OpenCV to avoid throttling.• Spearheaded a website for a school for the blind using Node.js, incorporating OCR via Google Vision API & LLM-powered Q&A with AWS Bedrock. Ensured accessibility with React component labeling, managed shared credentials via Firebase, & deployed on AWS EC2. Generated $120,000 in revenue within 1.5 months. Show less

• 

  KPMG India

  Mar 2024 - May 2024

  Summer Analyst (Data Science)

  • Created an LLM-based Database & Ledger Interaction System for SAP using Langchain, leveraging chain-of-thought prompting & instruction set fine-tuning of GPT4, achieving 76% accuracy in identifying data sources.• Built a Django-based web application extending SAP functionalities with the above-stated LLM Solution to process natural language prompts, fetch data & visualize, saving 5+ hours weekly for chartered accountants & 10+ hours for analysts.

• 

  Stony Brook University

  Aug 2024 - Dec 2024

  Graduate Teaching Assistant

  Teaching Assistant for IAE:101 (Digital Intelligence).