Iris: Centralize LLM model configuration and selection (#496)

Author: bassner

iris/COURSE_CHAT_PIPELINE_REFACTORING.md

@@ -193,7 +193,7 @@ their programming exercises on Artemis in a pedagogically meaningful way.
        - `api_key`: The API key for the model.
        - `description`: Additional information about the model.
        - `id`: Unique identifier for the model across all models.
-       - `model`: The official name of the model as used by the vendor, also used by ModelVersionRequestHandler for model selection (e.g., "gpt-4.1", "gpt-4.1-mini").
+       - `model`: The official name of the model as used by the vendor (e.g., "gpt-4.1", "gpt-4.1-mini").
        - `name`: A custom, human-readable name for the model.
        - `type`: The model type, used to select the appropriate client (e.g., `openai_chat`, `azure_chat`, `ollama`).
        - `endpoint`: The URL to connect to the model.
@@ -386,6 +386,45 @@ Deploying Pyris using Docker ensures a consistent environment and simplifies the
 
   Modify configuration files as needed:
   - **Pyris Configuration**: Update `application.yml` and `llm_config.yml`.
+    - `application.yml` must include an `llm_configuration` section that maps pipeline `implementation_id`s to variant IDs and roles with `local`/`cloud` LLM IDs (matching `llm_config.yml` `id`). Use `application.example.yml` as a template.
+    - Bulk-replace an LLM ID across all `llm_configuration` entries:
+
+      ```bash
+      # adjust these three values
+      export APP_YML=application.yml
+      export OLD_MODEL=gpt-4.1-mini
+      export NEW_MODEL=gpt-4o-mini
+
+      python - <<'PY'
+      import os
+      import pathlib
+      import yaml
+
+      path = pathlib.Path(os.environ["APP_YML"])
+      old = os.environ["OLD_MODEL"]
+      new = os.environ["NEW_MODEL"]
+
+      data = yaml.safe_load(path.read_text(encoding="utf-8"))
+      llm_cfg = data.get("llm_configuration", {})
+
+      def walk(node):
+          if isinstance(node, dict):
+              return {k: walk(v) for k, v in node.items()}
+          if isinstance(node, list):
+              return [walk(v) for v in node]
+          if isinstance(node, str) and node == old:
+              return new
+          return node
+
+      data["llm_configuration"] = walk(llm_cfg)
+      path.write_text(
+          yaml.safe_dump(data, sort_keys=False, allow_unicode=True),
+          encoding="utf-8",
+      )
+      print(f"Replaced {old} -> {new} in {path}")
+      PY
+      ```
+
   - **Weaviate Configuration**: Adjust settings in `weaviate.yml`.
   - **Nginx Configuration**: Modify Nginx settings in `nginx.yml` and related config files.
 

iris/README.MD iris/README.mdiris/README.MD renamed to iris/README.md

@@ -9,6 +9,27 @@ weaviate:
 memiris:
   enabled: true
   sleep_enabled: true
+  llm_configuration:
+    embeddings:
+      - mxbai-embed-large
+      - nomic-embed-text
+    learning_extractor:
+      local: gemma3
+      cloud: gemma3
+    learning_deduplicator:
+      local: gemma3
+      cloud: gemma3
+    memory_creator:
+      local: gpt-oss
+      cloud: gpt-oss
+    sleep_tool_llm:
+      local: gpt-oss
+      cloud: gpt-oss
+    sleep_json_llm:
+      local: gemma3
+      cloud: gemma3
+
+local_llm_enabled: true # Set to false to disable local LLM support entirely
 
 langfuse:
   enabled: false # Set to true to enable LangFuse tracing
@@ -30,3 +51,223 @@ transcription:
 
 env_vars:
   SOME: "value"
+
+# Values are LLM entry IDs (the "id" field in llm_config.yml).
+llm_configuration:
+  exercise_chat_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+    advanced:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  course_chat_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+    advanced:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  lecture_chat_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+    advanced:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  text_exercise_chat_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+    advanced:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  autonomous_tutor_pipeline:
+    default:
+      chat:
+        local: gpt-oss
+        cloud: oai-gpt-5-mini
+
+  tutor_suggestion_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+    advanced:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  competency_extraction_pipeline:
+    default:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  inconsistency_check_pipeline:
+    default:
+      solver:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      prettify:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+
+  rewriting_pipeline:
+    faq:
+      rewriting:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+      consistency:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+    problem_statement:
+      rewriting:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+      consistency:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+
+  lecture_unit_page_ingestion_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+    advanced:
+      chat:
+        local: oai-gpt-52
+        cloud: oai-gpt-52
+      embedding: azure-embedding-large
+
+  faq_ingestion_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+
+  transcription_ingestion_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+
+  lecture_unit_segment_summary_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+
+  lecture_unit_summary_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+
+  lecture_unit_pipeline:
+    default:
+      embedding: oai-embedding-small
+
+  lecture_retrieval_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+      reranker: cohere
+
+  lecture_unit_segment_retrieval_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+      reranker: cohere
+
+  lecture_transcriptions_retrieval_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+      reranker: cohere
+
+  faq_retrieval_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+
+  citation_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-nano
+      keyword_summary:
+        local: oai-gpt-5-nano
+        cloud: oai-gpt-5-nano
+    advanced:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+
+  interaction_suggestion_pipeline:
+    course:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-nano
+    exercise:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-nano
+
+  session_title_generation_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-nano
+
+  code_feedback_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-mini
+
+  summary_pipeline:
+    default:
+      chat:
+        local: oai-gpt-5-mini
+        cloud: oai-gpt-5-nano
+
+  lecture_search_answer_pipeline:
+    default:
+      hyde:
+        local: gpt-oss
+        cloud: oai-gpt-5-nano
+      answer:
+        local: gpt-oss
+        cloud: oai-gpt-5-mini
+      embedding: oai-embedding-small
+
+  mcq_generation_pipeline:
+    default:
+      chat:
+        local: gpt-oss
+        cloud: oai-gpt-5-nano