A single retrieval pass answers 'what is X' well. It fails on 'what changed between X and the update after it', because no single chunk contains both facts — the answer requires two lookups chained together.
Decompose, then retrieve twice
Have the model first break the question into sub-questions, retrieve for each separately, then answer using the combined context. This costs an extra LLM call but fixes a whole class of otherwise-unanswerable questions.
- Detect whether a question is likely multi-hop (comparisons, 'before/after', chained references).
- Decompose into 2–3 sub-questions with a cheap, fast model call.
- Retrieve separately for each sub-question, then merge contexts before the final answer.
Not every question needs this — it roughly doubles latency and cost, so gate it behind a cheap classifier rather than running it on every query. See query rewriting and HyDE for retrieval for a lighter-weight first step.
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.
