Metric AI Research Lab 👋

We are a frontier AI research lab based in Yerevan, Armenia.

We build and open-source models, datasets, and benchmarks in:

• 🤖 Physical AI & Robotics
• 📄 Document AI & Multimodal Retrieval
• 🌍 Low-Resource & Armenian AI
• 🧠 Multimodal Embeddings & Foundation Models

Selected research

Physical AI & Robotics

• FailBench — How Reliable Are VLMs at Judging Robot Task Success?
• A Statistical Audit of Physical AI Benchmark Redundancy
• [FailDetect — Robotic Failure Detection Framework]

Low-Resource & Armenian AI

• ArmBench-ASR — Armenian Speech Recognition Benchmark
• ArmBench-LLM — Armenian LLM Benchmark
• ATE-2 — Armenian Text Embeddings
• Less is More — EACL 2026 / LoResLM
• Uzbek RAG — EMNLP 2026 / Industry Track

Document AI

• ColQwen2.5 multilingual document retrievers
• ColQwenStella 2B visual document retriever