🥑
[rag]knowledge base········OK
[vec]bge-base · ONNX········OK
[llm]gemini-2.5-flash········LIVE
[sys]avocado online········✦ ready
>
Live observability
The dashboard an engineer would actually open: reliability and uptime, latency percentiles down to each pipeline stage, token cost, answer quality, and a live trace of recent requests. Avocado runs a hybrid retrieval pipeline — HyDE query expansion, dense (BAAI/bge-base) plus BM25 merged with Reciprocal Rank Fusion, a 1-hop knowledge graph, and a model-fallback chain. Everything here is read live from the production backend.