Jul 2024 – Apr 2025 · 9 mos · Chennai, Tamil Nadu, India · On-site

• Pioneered MILU, the first comprehensive benchmark for evaluating Large Language Models (LLMs) on authentic Indic contextual understanding
• Architected robust synthetic data generation pipelines for collecting high-quality audio data grounded in accurate Indian cultural and linguistic contexts
• Engineered a scalable, distributed translation infrastructure on Google Cloud Platform, processing millions of tokens to support the development of IndicTrans3 (Sarvam-M)
• Implemented advanced monitoring systems and automated job allocation services for the internal GPU cluster, optimizing resource utilization and computational efficiency
• Evaluation methodology and application development for IndicTrans3 and coordinated annotation teams to ensure quality assurance and consistent performance metrics

Feb 2023 – Jun 2024 · 1 yr 4 mos · Chennai, Tamil Nadu, India · On-site

• Worked on domain adaptation of Automatic Speech Recognition (ASR) systems using Class language models, training language models, evaluating and tuning hyperparameters of ASR models, generating and filtering data for training language models.
• Designed comprehensive end-to-end data scraping and filtering pipelines for rigorous evaluation of LLM capabilities and performance metrics
• Developed and successfully deployed critical internal data collection platforms, including the inaugural version of Anudesh and specialized annotation tools for comparative LLM output analysis for research projects.

Aug 2022 – Nov 2022 · 3 mos

• Expertly developed cutting-edge scripts and assessments, elevating Computer Vision standards.
• Coded and explained CV algorithms from scratch, such as CNN, through animations and relatable analogies
• Implementing and decoding state-of-the-art models like MobileNet, ResNet, EfficientNet, etc.
• Received 5/5 learner rating and appreciation from HOD Data Science.

Aug 2022 – Nov 2022 · 3 mos

Jun 2022 – Jul 2022 · 1 mo · Delhi, India

• Worked as a Summer ML Intern at the IITD AIA Foundation for Smart Manufacturing at IIT Delhi.
• Key Roles :
• Developing an end-to-end deep learning pipeline to identify power grid fault using voltage sensor data
• Researching and implementing state-of-the-art model architectures
• Deploying the Deep learning pipeline in Flask
• Hosting the Web App on Cloud to perform real time inference based on sensor data
• Optimizing the big data pipeline to minimize memory usage on training pipelines
• Delivered the deployed and hosted deep learning pipeline with ~85% accuracy

Jun 2022 – Jul 2022 · 1 mo · Delhi, India