Dec 2023 – Jun 2025 · 1 yr 6 mos · Los Angeles, California, United States · On-site

• Research Assistant at Language, Intelligence, and Model Ethics (LIME ) Lab
• Developed a framework for pre-hoc analysis of social intelligence of LLMs in gaming environments, enhancing understanding of intentions and impact on gameplay dynamics.
• Demonstrated gaps in LLMs with respect to intention inference (Theory of Mind) and self-regulation, highlighting areas for improvement in mimicking human-level social intelligence.
• Evaluating and mitigating hallucinations by Large Language Models in long form conversation.

Sep 2023 – Dec 2023 · 3 mos · Los Angeles, California, United States

• Research Assistant as part of DataFirst initiative.
• Received Best Data Science Developer Award
• Worked on multi-annotator models to analyze the relation between noise and annotator bias for subjective tasks.
• Implemented Dataset Cartography(uncertainty evaluation methodology) to evaluate the effect of annotator disagreement on prediction uncertainty.

Mar 2022 – Jul 2023 · 1 yr 4 mos · Bengaluru, Karnataka, India · On-site

• Developed a scalable and fault tolerant JDBC Source connector in MSK Connect for seamless and efficient data transfer from relational databases to Kafka.
• Implement an auto-scaling module in Spring Boot to scale the data transfer based on elastic search cluster health to avoid any back pressure.
• Implemented automated deployment pipelines, facilitating multi environment and region deployments to ECS, CodeArtifact, ECR, MSK and SQS AWS services.

Aug 2020 – Mar 2022 · 1 yr 7 mos · Bengaluru, Karnataka, India

• Trained regression models on turnaround data of vehicles from Sixt branches to predict vehicle servicing times and built an active feedback training loop for the model to learn on the fly in production.
• Designed and implemented modules for 9 microservices in Operations app in SpringBoot and Go-Micro, enabling migration to a distributed architecture.
• Worked on optimizing the system through indexing, analysis of slow queries and subsequent improvements and scheduled archival of tables.

Jan 2020 – Apr 2020 · 3 mos · Bengaluru Area, India

• Developed a fuel receipt scanner using MLKit and OpenCV, predicting with more than 99% accuracy the refueled amount and price through document detection in view finder and OCR on images.
• Improved the Check-in process reliability in Operations App with development in Spring Boot, reducing error rate from 12% to 0.5% through bug identification, fixes, and resolving cross-system synchronization errors.

Jan 2019 – Mar 2019 · 2 mos · Bengaluru Area, India

• The project involved porting the current website of Learnyst to a static CMS using Jekyll. The role also involved making the website editable on the go by the Learnyst marketing team and improving the SEO score of the website.

Jun 2018 – Jul 2018 · 1 mo · Bangalore · On-site

• • Fine-tuned a Tacotron2 model on internal dataset recordings for voice assistant platform, on text embeddings and Mel-spectrograms. Enhanced speech synthesis quality, particularly for domain-specific technical terms.