{"id":"ai-ml-stack","type":"module","title":"AI/ML stack at a glance","status":"ga","audience":["engineer","analyst","prospect"],"scope":"both","tier_min":"free","aliases":["appendix","models"],"summary":"Consolidated reference of every model and service the platform uses, what each is for and where. Providers marked currently configured are env-pluggable through the LiteLLM factory.","keywords":["litellm","bedrock","qwen","claude","cohere","bge-m3","pgvector","mephiston","reranker","dspy"],"body":{"what_it_is":"Consolidated reference of every model/service mentioned across modules.\nProviders and models marked \"currently configured\" are env-pluggable via the\nLiteLLM factory; dev defaults shown, production paths noted.\n","what_it_helps_users_do":null},"engines":[{"id":"litellm-factory","type":"engine","title":"LiteLLM client and factory","summary":"The provider abstraction every gateway-side LLM call goes through. Selects Bedrock, Azure OpenAI, OpenAI or Gemini, records per-call cost, latency and tokens, and stamps a prompt hash for replay.","category":null},{"id":"embedding-service","type":"engine","title":"Embedding service","summary":"Produces 1024-dim vectors from Cohere Embed v4, OpenAI or bge-m3, all pinned to the same dimensionality so the vectors are interchangeable across providers.","category":null},{"id":"mephiston-vector-db","type":"engine","title":"Mephiston vector database","summary":"A Rust hybrid dense-and-sparse chunk-level store with tenant isolation, sector and geo filters and a circuit breaker that falls back to Postgres.","category":null},{"id":"g3-extractor","type":"engine","title":"G3 requirement extractor","summary":"Extracts a tender's individual requirements as claims, with a hallucination harness that requires every item to carry a verified verbatim evidence span.","category":null}],"integrations":[],"capabilities":[]}