A student asked why their RAG bot kept citing a 2022 policy document when a 2025 update existed. Both were in the index, both matched semantically, and vector search had no idea one was stale.
Filter before you search, not after
Store metadata at ingestion — document date, source, section type, access level — and apply it as a pre-filter in the vector query, not as a post-hoc check on the results. Every major vector store supports this natively.
- Date: filter to the latest version of a document family, or let the user pick a range.
- Source: separate official docs from forum answers so one never silently outranks the other.
- Section type: a table of contents chunk should not compete with body-text chunks for the same query.
The failure mode this prevents
Without filtering, semantic similarity is the only signal, and a well-written old answer beats a plainly-written current one. Metadata filtering fixes this before reranking ever runs, which is also cheaper — fewer candidates to rerank.
Pairs well with hybrid retrieval: dense, BM25 and RRF once the candidate set is already clean.
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU Kanpur) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. Email pranjulrathour41@gmail.com.
