Skip to content

Interview Preparation

Two-minute script

“Vanijya AI is a trade-intelligence platform that unifies structured trade analytics with document-grounded policy search. The RAG service ingests difficult PDFs and tables, stores originals in MinIO, metadata/history in Postgres, dense vectors in Qdrant, sparse terms in Meilisearch, and sliding-window memory in Redis. Queries can be rewritten or decomposed, retrieved through hybrid search and reranking, and streamed with citations and debug traces. Dedicated project scopes support FTA, HS, confidential dossier, and concordance use cases.”

Likely questions

  • Why combine Qdrant and Meilisearch?
  • How do project IDs prevent confidential document leakage?
  • When is model fallback acceptable for HS/concordance?
  • Why keep Redis separate from durable history?
  • How do you observe retrieval quality and latency?
  • How do structured and unstructured workspaces share SSO?

Tradeoffs

Hybrid retrieval adds operational complexity but handles both semantic policy language and exact tariff codes. Rich document extraction improves answer quality but increases CPU, dependency, and failure surface. Project-specific prompts improve relevance but require careful configuration governance.