Joining Uber's engineering team as a full-time software engineer.

Built streaming data pipelines (Kafka, FastAPI, asyncio) processing 1M+ financial records/day. Developed RESTful APIs reducing response latency by 60%. Engineered MongoDB feature store restoring sub-second query performance on millions of records.

Developed Python ETL pipeline processing 500+ SEC filings per batch. Integrated GPT-based NLP achieving 99% entity extraction accuracy. Built Flask + AWS (EC2, RDS) REST API serving 50+ QPS with p95 latency <200ms.

Building LLM-based file reorganization agents and VLM-augmented slide QA systems for the Berkeley Teaching AI platform.

Co-designed Matryoshka, an ML-centric dataset discovery and feature selection system scaling to 100M+ row data lakes using DuckDB and PostgreSQL.

Trained ensemble ML models (Ridge, LSTM) on multi-terabyte ABCD datasets for longitudinal adolescent health prediction; achieved AUC >0.80.

Built ML forecasting platform processing 50M+ energy rows across 22 provinces; NLP pipeline analyzing 1,000+ policy documents for China's mid-century power analysis.

Co-developed Re-NeRF, optimizing high-resolution NeRF training with deformable convolution; improved rendering quality in 4K scenes (PSNR/SSIM).