May 2025 – Aug 2025 · 3 mos

• Contributed to the Garg Research Group by leading the software architecture and full-stack development of an IoT-enabled cement strength testing device, working with a cross-functional team of engineers and researchers to modernise a critical operation in the construction materials industry. The goal is to design scalable, automated, and data-driven systems that seamlessly connect embedded hardware with cloud applications.
• Responsibilities:
• 1) Lead the end-to-end design and engineering of a full-stack web application using React, Node.js, Express, and PostgreSQL, applying modern SDLC practices to ensure scalability, maintainability, and production readiness. Also contributed to the PyQt5 legacy desktop app, maintaining compatibility across platforms.
• 2) Enable real-time device-to-cloud connectivity by leveraging the MQTT protocol (HiveMQ Cloud broker), creating a reliable, low-latency communication pipeline between embedded hardware and the web dashboard accessible globally.
• 3) Implement networking infrastructure on Raspberry Pi via RaspAP Auto Access Point mode, allowing seamless plug-and-play device setup and wireless connectivity without requiring external configuration.
• 4) Develop multi-stage execution pipelines that coordinate low-level hardware instructions on Raspberry Pi with the web application, ensuring synchronisation between device tasks, user inputs, and backend services.
• 5) Integrate computer vision pipelines using OpenCV to automatically extract regions of interest (ROI) from cuvette images, calculate concentration and dissolution indices, and deliver real-time analytics to researchers through the dashboard.
• I collaborated in a cross-disciplinary environment (Electronics, Mechanical, and Researchers), translating research needs into software requirements and helping interview, onboard, and mentor new contributors joining the project.
• Skills: Full-Stack Development, Automation, Cross-Functional Collaboration, Technical Mentorship, Leadership.

Feb 2025 – May 2025 · 3 mos

• Assisted Prof. Matthew Caesar in CS437: Internet of Things, focusing on evaluating and improving student projects in embedded systems, networking, and IoT application development.
• Responsibilities:
• 1) Grade projects/machine problems related to IoT protocols, distributed systems, and edge-device integration, ensuring technical accuracy and adherence to best practices.
• 2) Provide detailed technical feedback to students on design, implementation, and debugging, strengthening their skills in embedded programming, data pipelines, and real-time systems.
• 3) Collaborate with TAs and faculty to refine course content, ensuring assignments reflect industry-relevant technologies.
• Skills: IoT Systems, Code Review, Software Systems, Technical Communication, Mentorship, Collaboration, Teamwork, Excel.

Aug 2024 – Dec 2024 · 4 mos · Champaign, Illinois, United States · On-site

• A side hustle that ultimately strengthened time-management, teamwork, multicultural and communication skills in a fast-paced environment.
• Delivered friendly and efficient customer service to hundreds of students daily during high-traffic dining hours.
• Collaborated with a diverse team to support meal preparation, serving, and inventory tasks.
• Received commendation from the manager for going above and beyond duties to better serve people.

Aug 2023 – Jul 2024 · 11 mos · Mumbai, Maharashtra, India · On-site

• Engineered a Python-based, multithreaded ETL + forecasting tool, processing up to 8 million SAP HANA records monthly for an electronic components manufacturer and supplier client, achieving a 10% reduction in MAPE and 3× faster SKU-level time-series modelling.
• Automated and orchestrated production ETL workflows with modern CI/CD and scheduling tools (GitHub-like SVN, Jenkins, Cron,), ensuring reliable monthly runs and reducing manual effort by 50%.
• Built and integrated data ingestion and validation modules for external regressors from the FRED API, uncovering key demand drivers and improving forecast reliability by 20%.
• Documented the pipeline architecture and data flow to support scalability, onboarding, and long-term maintainability.
• Continuous experimentation and benchmarking of the state-of-the-art models with respect to the deployed model.
• Collaborated with the Data Science team and Domain experts to investigate the gaps and opportunities in the purchase trends for the client and provide appropriate advice to the stakeholders.

Aug 2022 – Jun 2023 · 10 mos · Pune, Maharashtra, India · Hybrid

• Co-authored 3 publications on domain-adaptive pretraining, datasets, and transformer-based NLP libraries (Springer/AACL)
• Research Motivation:
• High-quality data is crucial for effective machine learning models. For LLMs, pretraining on large unstructured data can be tedious. Our research explored multiple data selection strategies to extract in-domain corpus from generic-domain data, enhancing model alignment with domain knowledge.
• Key Contributions:
• Developed - mahaNLP, an open-source Marathi NLP toolkit built on MahaBERT-based transformer models, significantly improving Marathi text analysis. Enabled researchers and developers to process 100,000+ Marathi text datasets with state-of-the-art accuracy, advancing NLP for low-resource Indian languages. Published in IJCNLP 2023
• Curated L3Cube-MahaNews, the largest supervised Marathi text classification dataset (108,643 records across 12 categories), facilitating document-length analysis with advanced BERT models. Published in SPELLL 2023
• Designed 'TextGram', a rank-based domain-adaptive data selection strategy for LLM pretraining, achieving a 1% increase in F1-score on IMDb classification and outperforming existing techniques on the RealNews dataset. Published in SPELLL 2023
• Conducted extensive literature reviews on advanced NLP techniques, driving innovation in low-resource language processing.

Jun 2022 – Aug 2022 · 2 mos · Pune, Maharashtra, India · Remote

• Key Contributions:
• Developed an end-to-end retrieval pipeline with intelligent contextual search, leveraging the Haystack QA system to efficiently extract information from policy documents.
• Conducted Vulnerability Assessment and Penetration Testing (VAPT) to identify and mitigate security issues, enhancing application security and robustness.

Mar 2022 – May 2022 · 2 mos · Pune

• Worked on the task of Optical Handwritten Charater Recongition under Dr. Geetanjali Kale to automate the task of checking the answer sheets written by students during examination which is generally done by human examiner.
• The dataset was manually curated with in-college exam sheets database.