Candidate sourcing for recruiters is best viewed as a two-stage retrieval and reranking pipeline with recall as the primary objective under a limited review budget. In the primary setting studied in this paper, an upstream production retriever first returns a candidate shortlist for each job description (JD), and our goal is to rerank that shortlist so that qualified candidates appear as high as possible. We present mira-embeddings-v1, a semantic reranking system for the recruitment domain that reshapes the embedding space with large language model (LLM)-synthesized training data and further corrects boundary confusions with a lightweight reranking head.
Starting from real JDs, we build a five-stage prompt pipeline to generate diverse positive and hard-negative samples that sculpt the model's semantic space from multiple angles. We then apply a two-round LoRA adaptation: JD–JD contrastive training followed by JD–CV triplet alignment on a heterogeneous text JD–CV dataset. Importantly, these gains are pursued without large-scale manually labeled industrial training pairs: a modest set of real JDs is expanded into supervision at the task level through LLM synthesis. Finally, a BoundaryHead multi-layer perceptron (MLP) reranks the Top-K results to further distinguish between roles that share the same title but differ in scope.
On a local pool of 300 real JDs with candidates returned by an upstream production retriever, mira-embeddings-v1 improves Recall@50 from 68.89% (baseline) to 77.55% while lifting Precision@10 from 35.77% to 39.62%. On a supportive global pool protocol over 44,138 candidates judged by a Qwen3-32B rubric, it achieves Recall@200 of 0.7047 versus 0.5969 for the baseline. These results show that supervision synthesized by LLMs with data efficiency, plus reranking with boundary awareness, yields robust gains without introducing a heavy cross-encoder in our current pipeline.