Contrastive fine-tuning of a single embedding model across heterogeneous retrieval tasks is challenging when the tasks differ in label structure. In product-compliance retrieval, some tasks have multi-label supervision — every item carries a set of compliance categories, and many in-batch items share at least one category with the query, producing frequent in-batch false negatives. Other tasks (e.g., cross-catalog product matching) have no category labels: each query has one canonical match. Existing multi-label contrastive methods require every batch item to carry comparable labels, so they cannot serve a single embedding model trained on both label-bearing and label-free tasks. We propose label-aware false-negative exclusion: a zero-cost binary mask on the InfoNCE denominator that drops same-label candidates from the negative set rather than including them to positives. The mask has no effect on label-free items, so the same loss handles both task types. A 0.6 B-parameter embedding model fine-tuned this way handles four retrieval tasks simultaneously, lifting average Recall by 49% (0.489 → 0.729) and matching the 13× larger off-the-shelf Qwen3-Embedding-8B at 32× lower embedding dimensionality. Fine-tuning further enables learned query fusion: concatenating regulatory text with an example product, which hurts retrieval before fine-tuning, becomes the strongest query type after, beating either alone by 6-11%.

E-commerce faces persistent challenges with data quality issue of product listings. Recent advances in Large Language Models (LLMs) offer a promising avenue for automated product listing enrichment. However, LLMs are prone to hallucinations, which we define as the generation of content that is unfaithful to the source input. This poses significant risks in customer-facing applications. Hallucination detection is particularly challenging in the vast e-commerce domain, where billions of products are sold. In this paper, we propose a two-phase approach for detecting hallucinations in LLM-enriched product listings. The first phase prioritizes recall through cost-effective unsupervised techniques. The second phase maximizes precision by leveraging LLMs to validate candidate hallucinations detected in phase one. The first phase significantly reduces the inference space and enables the resource-intensive methods in the second phase to scale effectively. Experiments on two real-world datasets demonstrated that our approach achieved satisfactory recall on unstructured product attributes with suboptimal precision, primarily due to the inherent ambiguity of unstructured attributes and the presence of common sense reasoning. This highlights the necessity for a refined approach to distinguish between common sense and hallucination. On structured attributes with clearly de- fined hallucinations, our approach effectively detected hallucinations with precision and recall surpassing targeted level.