msu-denver/bili-core

bili-core is a thin orchestration layer built on well-established, commodity components (LangChain, LangGraph, Streamlit). It wraps existing LLM evaluation patterns without introducing novel benchmarking methodology, metrics, or architectural innovation. The 11 stars, zero velocity, and 421-day dormancy indicate minimal adoption and community traction—classic hallmarks of an academic/research project that hasn't achieved product-market fit. The framework offers standard functionality: multi-model comparison, RAG testing, and decision workflows—all capabilities that frontier labs (OpenAI, Anthropic, Google) are actively embedding into their own platforms (Anthropic's evals, OpenAI's batch APIs, Google's Vertex AI evaluators). The project does not define a standard, owns no unique dataset, and lacks switching costs. A user could equally use LangSmith (Anthropic's native benchmarking), LiteLLM's evals, or write custom LangChain code. The MSU Denver context positions this as an academic toolkit, which further reduces defensibility against platform consolidation. No README evidence of novel evaluation metrics, domain-specific RAG testing, or workflow patterns that couldn't be replicated in a few hours by someone familiar with LangChain. High frontier risk because this is exactly the kind of glue-code that frontier labs subsume as they mature their evaluation and orchestration tooling.