System Report · July 2026 Mira: The First End-to-End AI Recruiter
The Metix AI Team
How Mira is designed from the ground up: five specialized sub-agents, a talent graph of over one billion public profiles, and two in-house fine-tuned models that turn hiring intent into scheduled interviews.
A recruiting-native embedding model lifted our primary business metric over 40% and cut operational cost per qualified, interested candidate more than 90% versus a general-purpose baseline.
Read research → Engineering Note · July 2026 Agent Evaluation, Done Right
Zhilin Wang
Why the durability of a production agent depends on its evaluation system: research-style testing, layered scoring, golden sets, calibrated judges, and evaluation wired into CI.
Bug-free code and correct agent behavior are two separate claims; what keeps a production agent standing is the evaluation system behind it.
Read research → Engineering Note · June 2026 Performance Drift in Agent Systems
Zhilin Wang
A structural look at why production agent systems drift across the prompt, architecture, evaluation, and context layers, even when the spec and business goal stay fixed.
Agent drift is structural, not incidental: teams need prompt versioning, layered evaluation, golden sets, observability, context engineering, and deterministic fallbacks.
Read research → arXiv Preprint · April 2026 Mira-Embeddings-V1: Domain-Adapted Semantic Reranking for Recruitment via LLM-Synthesized Data
Zhaohua Liang, Zhilin Wang, Renjie Cao, Yining Zhang
A recruitment-domain semantic reranking system that uses LLM-synthesized supervision and boundary-aware reranking to improve candidate retrieval recall.
Recall@50 improved from 68.89% to 77.55% on a local pool built from 300 real job descriptions.
Read research → OpenJobs AI Review Paper · March 2026 When AI Meets Recruiting: Opportunities, Challenges, and Future Directions
Yining Zhang, Renjie Cao, Zhilin Wang
A lifecycle-oriented review of AI recruiting systems, covering semantic matching, generative AI, multimodal assessment, bias, explainability, and human oversight.
Recruitment AI is moving from isolated prediction tasks toward lifecycle-oriented, generative workflows.
Read research →