Applied AI Research Engineer, Starburst

• 3+ years of experience in information retrieval, NLP, knowledge representation, or applied ML research
• Production experience building RAG, grounding, or retrieval systems (not prototypes or demos)
• Strong evaluation methodology: benchmark design, statistical analysis, reproducible experiments
• Comfort operating at the research/systems boundary: you read papers and you ship code
• Python fluency; experience with vector databases, embedding models, LLM APIs
• Track record of converting research insights into shipped production systems

Desirable

• We build the AI layer for Starburst's products, including AIDA. We design agents that let users ask questions in natural language and get accurate, grounded answers backed by their actual data. We operate with startup speed inside an enterprise company, shipping weekly and measuring results. This is the first dedicated research engineering hire on the team
• You will own the intelligence layer that makes AIDA's agents correct, trustworthy, and measurably better over time
• The work spans information retrieval, knowledge representation, and evaluation science. You will turn ambiguous notions of "agent quality" into clear metrics, build the grounding systems that connect agent reasoning to verified data, and create the evaluation infrastructure that makes quality a first-class engineering discipline
• You will operate at the research/systems boundary: running experiments with academic rigor and shipping results with production engineering discipline. Research and engineering are not separate tracks here. You will own experiments end to end, from hypothesis through production deployment
• Design and build grounding systems that connect agent reasoning to verified enterprise data sources
• Build and optimize retrieval pipelines (RAG, hybrid search, structured query generation) for accuracy and latency
• Define data representation strategies that preserve semantic fidelity across heterogeneous enterprise data (catalogs, schemas, lineage)
• Create evaluation frameworks: automated benchmarks, regression suites, human evaluation protocols
• Convert validated research findings into production systems that ship to users
• Establish quality metrics and dashboards that track agent correctness week over week
• Build feedback loops where user interaction data flows back into evaluation datasets and informs grounding improvements