Kunal Raha

@moeed_chughtai Hey Moeed — the context reconstruction problem is real. The hardest technical challenge here is probably relevance ranking when a lawyer queries across hundreds of documents. Curious — how are you handling semantic search? Pure vector similarity or are you using a reranker on top to improve precision? BGE-M3 cross-encoder reranking could significantly improve your retrieval...