Building and scaling a real-time audio streaming and transcription platform, architecting multi-language ASR pipelines across globally distributed environments.

Driving end-to-end platform reliability while leveraging LLMs to generate AI-powered meeting artifacts, enabling seamless real-time speech-to-text and intelligent meeting insights for millions of users.

Building the recommender system that powers the Twitter Home Timeline, the largest scale and most impactful consumer product at Twitter (> 75% of time on Twitter is spent on Home).

Experienced in building scalable and fault-tolerant distributed systems. Shipped end-to-end improvements to the entire online and offline codepath covering critical online serving infrastructure, feature engineering, performant data pipeline and running advanced A/B tests at scale.

Optimized online serving path component with SLO of 10M predictions/sec, 50M QPS. Improved service tail latencies by > 20% and achieved massive reduction in ranking failures. Improved service scalability, reliability, and maintainability.

Redesigned streaming and offline data pipelines to increase production training data collection by 30, O(100TB), while removing training/inference discrepancy. This resulted in significant model and ranking quality gains on Twitter’s largest consumer facing product surface: +16M UAM, +500k unique engagers.

Migrated core service components out of a legacy codebase into a more performant framework.

Keywords: distributed systems, backend, ML, Python, Java, Scala, AWS, GCP, BigQuery, SQL, ETL, Kafka, Hadoop

Interned at Uber Advanced Technology Group (Uber’s self-driving unit, later acquired by Aurora Innovation).

Designed and deployed cloud-based end-to-end CI/CD log classification system, automated services setup with Terraform modules.

Implemented data pipeline with AWS services, engineered error log classification with Spark and Python.

Wrote SQL queries to interact with large data systems.

Released real-time visualizations metrics, detecting infra failure vs. human errors.

Expedited infra failure detection process, improved developer productivity.

Keywords: backend, CI/CD, AWS, Spark, Python, SQL, databases, Hive, Terraform

Led office hours to present and review material taught by the professor in class with 40 students.

Assisted students in understanding learning material including feature detection, image segmentation, motion estimation, object recognition, and 3D shape reconstruction through stereo, photometric stereo, and structure from motion.

Processed large-scale human genome data using Python and shell commands over high-performance-computing clusters.

Performed data visualization and exploratory statistical analysis using R and Python.

Built and assessed machine learning model that predicts the specific biological effect of regulatory variants in the context of gene expression measured in cells or tissues, which offers insights into whether a gene variant is disease-causing or not.

Keywords: data science, statistical analysis, machine learning, Python, R, Unix commands, git