Jul 2025 – Oct 2025 · 3 mos · Bengaluru, Karnataka, India

• TableTorch (https://www.tabletorch.ai/): State of the art tabular data extraction software - 96.7% accuracy on the Nanonets IDP benchmark. Adaptations to make the model run inference on CPU with a P90 latency of 0.2 second per page.
• ML Engineering Consultant, Sapient AI: Designed and implemented automated grading systems for medical education programs. Architected a spreadsheet agent that reduced case-to-assignment workflows from hours to under 5 minutes, improving operational efficiency for medical school instructors.
• Mathematics Tutor (Volunteer): Volunteered as a remote mathematics tutor for 8th grade students at a school in rural Uttar Pradesh.
• Started my fledgling youtube channel (https://www.youtube.com/@vyom_thakkar) where I share what I'm learning and create content on topics (ML, RL, systems programming) that I personally struggled with.
• Started learning Mandarin.

Aug 2022 – Jun 2025 · 2 yrs 10 mos · Hyderabad, Telangana, India

• Drove end-to-end responsibilities across Backend Development, Machine Learning, Signal Processing and Cloud Infrastructure
• Developed robust motion artifact detection and denoising algorithms by researching, implementing, and optimizing state-of-the-art digital signal processing techniques for Photoplethysmography (PPG) signals.
• Enhanced existing data pipelines through optimization and caching strategies; redesigned high-latency microservices, achieving a 53% reduction in computation time.
• Led the development of an LLM-powered health analyst agent, enabling users to gain comprehensive insights and identify patterns within their health data, utilizing scalable and reliable agent-based frameworks and best practices.
• Led the development of a compute-intensive “Trends” feature, enabling users to analyze their data across multiple time scales; implemented a time-interval based hashing datastructure that sped up API calls to tens of milliseconds per request, bringing down latency by 400x.
• Built a simulation environment enabling rapid backtesting of core algorithm changes against historical data, significantly accelerating analysis and validation of model improvements.
• Implemented server-side functionality of an end-to-end D2C marketplace (on app and website) that supports payments and automates logistics for order fulfillment.

Jan 2021 – May 2022 · 1 yr 4 mos · Champaign, Illinois, United States

• Worked on an NLP-Powered Web App (MERN stack) that performs Automated Short Answer Grading of medical chart notes with the aim of providing real-time feedback to medical students.
• Designed and implemented the full-stack lifecycle of the platform and integrated end-to-end data flow pipeline.
• Developed student-view of the React frontend and styled using Tailwind CSS.
• Implemented the Node.js backend along with user-authentication in Passport.js.
• Automated data collection & processing pipeline for the NLP system, that reduced time-to-view grades by 10x.
• Improved the accuracy of NLP model by 16% by making use of AWS Comprehend for medical entity extraction and implementing BioWordVec clustering.
• The system was piloted with 5 cohorts of medical students and presented during the International Meeting for Simulation in Healthcare in Los Angeles, Jan 2022.

Jan 2020 – Dec 2020 · 11 mos · Champaign, Illinois, United States

• Implemented NLP software to detect help-seeking student posts in large course discussion forums, with the aim of simplifying instructors’ task in locating and answering these questions.
• Experimented with different NLP models including LSTM, Naive Bayes, Logistic Regression, Random Forest Classifiers and obtained testing accuracy of 89%.
• Expanded the training dataset by mining text data from Reddit using the Reddit REST API as well as the Stanford MOOC dataset and adapted the data to the help-seeking classification task.

Jun 2018 – Aug 2018 · 2 mos · Urbana, Illinois, United States

• • Refactored the backend code, implemented unit-testing suite and stress-tested the C2E2 verification tool (for cyberphysical systems) in different simulation environments.

Oct 2017 – May 2018 · 7 mos

Jun 2017 – Aug 2017 · 2 mos · Mumbai Metropolitan Region

• Developed software to compute the angle between the tibia and femur bones for a computer-assisted surgery system.
• Implemented code to transfer real-time coordinate data between the hardware and software via UART protocol.