2026-11-03 –, Seminar Room 2
The ground system for NASA's NEO Surveyor mission at Caltech/IPAC uses a heavily partitioned PostgreSQL database and Python tooling. This poster describes practical experience using large language models for code review, library migrations, schema and query design, and moving application database access to a stored-procedure API. The engineer defines the task, selects relevant context, and verifies each proposal. Controlled tests showed that AI can detect problems in application code but can also produce incorrect results when important context is missing. The poster presents a workflow in which AI drafts solutions, tests provide evidence, and the engineer decides what to implement.
Tatiana Mosheva is a database engineer at Caltech/IPAC, working on NASA’s Near-Earth Object Surveyor (NEO Surveyor) mission. Her work includes PostgreSQL schema design, partitioning, query APIs, and Python tooling for testing and operations. She also uses AI as an engineering assistant, with one small condition: “sounds convincing” does not count as a passing test result.