Context
Worked on NLP, information retrieval, ranking, and large-scale text systems at USC Information Sciences Institute with leading researchers across search, summarization, and data integration.
What I built
- A real-time Twitter search system used in DHS projects and awarded "Best System" at the 2011 TREC Microblog Track
- GeoXray, a geographic text summarization system later acquired by TerraGo
- Large-scale biomedical data integration systems across six medical centers for NIH research
- Co-authored research proposals and 10+ papers spanning microtext search, geographic summarization, and data integration
What was technically hard
- Building real-time search over the Twitter firehose at a time when the tools for processing social media at scale barely existed
- Designing geographic summarization systems that could extract meaningful spatial patterns from unstructured text
- Integrating biomedical data across six medical centers with different schemas, standards, and data quality levels
What I learned
- How NLP, ranking, and retrieval systems work at a foundational level, not just as APIs
- How to build research prototypes that become production-oriented systems
- How to collaborate with strong scientists and turn research ideas into working software
- The discipline of publishing, proposing, and defending technical ideas
Why it matters now
This period formed the technical foundation for everything that followed. The AI foundation is not recent hype. It came from years of serious NLP and systems research. That depth shows up in how I evaluate AI products, review model architectures, and help founders distinguish real technical moats from marketing.