EXPERIENCE
Research Assistant - Whitebox AI Lab
- Working under Prof. Abhinav Verma to reproduce the ArtiSAN Deep Reinforcement Learning framework to optimize atomic structures in High-Entropy Alloys.
- Investigating novel reward functions to incorporate configurational entropy for Gibbs Free Energy minimization, while leveraging AlphaFold-inspired mechanisms to interpret and enhance atomic swapping logic.
Deep RLAlphaFoldPhysics-MLResearch
NLP Intern
Titan Company Limited
- Built a real-time voice agent using OpenAI's Realtime API and a custom FastAPI server for conversational AI.
- Integrated Twilio & WebSockets to stream low-latency, bi-directional audio to both landline and mobile phone networks.
- Built a RAG pipeline using LangChain & a Redis vector database to provide the agent with context-aware responses.
- Developed interruption handling and semantic caching to enhance conversational flow and significantly reduce API latency.
OpenAI Realtime APIFastAPITwilioRAGRedis
Data Science Intern
Dr. Reddy's Laboratories
- Developed a data pipeline using Python & SQL to calculate the 'cost of poor execution' (COPE) by performing ETL on Google BigQuery data.
- Analyzed COPE distribution to identify key risk segments, saving an estimated 180+ hours of manual effort annually.
- Leveraged an LLM to accelerate building a multi-view dashboard in React & JavaScript for visualizing COPE metrics.
- Optimized legacy data analysis code for historical inventory tracking by implementing Dask for parallel processing.
PythonSQLBigQueryLLMReactDask
Research Assistant - NLP Lab
- Worked under Dr. Wenpeng Yin on a study investigating LLM 'blind compliance' versus metacognitive reasoning, evaluating state-of-the-art models' ability to identify logically flawed premises.
- Curated and annotated a specialized dataset of erroneous questions, categorizing logical fallacies to benchmark model robustness and define failure modes in zero-shot reasoning tasks.
NLPLLM ResearchBenchmarkingDataset Creation
Research Intern
BITS Pilani
- Developed a modular PyTorch framework to optimize an image classifier via a comparative analysis of CNN architectures.
- Implemented Tensor Methods for lossless compression on a dataset of 1.5 million high-resolution images.
- Deployed the image classifier into a production app, achieving 98% validation accuracy in predicting structural damage.
PyTorchCNNTensor MethodsComputer Vision
Data Analyst
The Daily Collegian
- Automated a data pipeline using GraphAPI, Express, and Axios to track user engagement statistics for a Facebook page.
- Derived actionable insights from engagement metrics & demographics, boosting engagement by 15% in 4 months.
GraphAPIExpressData AnalyticsAutomation
Research Assistant - HTI Lab
- Worked with Dr. Yiqi Zhang on analyzing driver takeover quality in conditionally autonomous vehicles using eye gaze data.
- Built an XGBoost classification model using eye gaze & simulator data to predict driver takeover in AVs with 94% accuracy.
- Analyzed takeover quality vs. interface design using Tobii2 gaze data; presented findings at a university research exhibition.
XGBoostMachine LearningEye TrackingHCI
Learning Assistant
Penn State College of Engineering
- Worked with instructors to create meaningful assignments and course work for students in a 200 level course (Object-Oriented Programming with Java).
- Responsible for assisting with in-class activities, feedback on assignments, grading, and tutoring.
JavaOOPTeachingMentorship
Research Assistant - HTI Lab
- Worked with Dr. Yiqi Zhang's group to design various decision-based interaction scenarios between Human driven and Autonomous vehicles.
- Augmented the behavior of autonomous vehicles in STISIM using SDL (Scenario Design Language).
- Used Action Trigger Settings, Position Settings and Basic Physics to match the constraints set forth in the decision-based interaction scenarios.
STISIMSDLAutonomous VehiclesSimulation
Treasurer
Indian Cultural and Language Club
- Spearheaded fundraising and organized events for over 10,000 attendees, generating event revenues of $30,000.
LeadershipFinanceEvent Management