Atita Arora is the founder and principal consultant of Bistro Innovation Labs, a Berlin-based advisory focused on search, retrieval-augmented generation, and enterprise AI. Her work brings established information-retrieval disciplines—relevance judgments, domain-specific ranking, and systematic evaluation—to systems increasingly responsible for generating answers and taking actions.
Arora began working in search at Amsoft Systems in 2007, integrating Apache Lucene and Solr. She subsequently worked across e-commerce and enterprise search, including at myToys and Lucidworks, before becoming a search relevance consultant at OpenSource Connections. There, she helped implement vector search in Chorus, an open-source e-commerce search platform, and documented how semantic retrieval integrates with Solr, product catalogs, query rewriting, and relevance assessment. She also contributed to Quepid relevance evaluation, including methods for comparing search-result sets.
At Qdrant, Arora worked as a solutions architect and developer-relations manager, applying her search background to vector databases and evaluation-driven RAG optimization. Her writing on search and retrieval challenges and AI Engineer World’s Fair appearance with Deanna Emery examined how document extraction, chunking, embeddings, metadata, retrieval, and context selection collectively determine answer quality.
She subsequently became director of AI at Voyager Search, working on a multimodal enterprise knowledge assistant incorporating optical character recognition, vision-language models, and secure, air-gapped retrieval. She founded Bistro Innovation Labs in 2025.
- Relevance is domain-specific. Useful ranking depends on what users must accomplish: product availability and compatibility in commerce, specialized terminology in technical search, or location in geospatial systems.
- Evaluation should precede complexity. Start with a straightforward retrieval pipeline, locate failures in extraction, chunking, retrieval, or generation, and introduce reranking or query rewriting only when measurements justify it.
- Hybrid retrieval combines meaning and precision. Dense vectors capture semantic relationships; sparse retrieval preserves exact matches for identifiers, acronyms, and technical vocabulary.
- Agent reliability depends on retrieval quality. When AI systems take action, selecting the wrong information can produce consequential downstream mistakes.
Arora also co-leads Women in Search, supporting the professional community around information retrieval.
