feat: Consolidate on a single LangGraph orchestration (#540)

This PR removes LangChain tools and Vertex AI Function Calling
orchestration and consolidates on LangGraph.

It also flattens the directory structure and refactors to simplify some
parts of code for easier understanding.

> [!NOTE]
> The failure in the integration test is expected. This PR is part of a
series of changes, and the corresponding fix for this test is in a
subsequent PR.
> ### Reasoning
> We've intentionally split the work into smaller, focused PRs to make
the review process more manageable and efficient.
> ### Merge Plan
> All related PRs will be merged into the `toolbox-main` branch first.
We will ensure all tests are passing on `toolbox-main` before merging
the entire feature set into `main`.

Author: anubhav756

.github/sync-repo-settings.yaml

@@ -35,9 +35,7 @@ branchProtectionRules:
       - "retrieval-service-postgres-pr (retrieval-app-testing)"
       - "retrieval-service-alloydb-pr (retrieval-app-testing)"
       - "retrieval-service-cloudsql-pg-pr (retrieval-app-testing)"
-      - "llm-demo-langchain-tools-pr (retrieval-app-testing)"
       - "llm-demo-langgraph-pr (retrieval-app-testing)"
-      - "llm-demo-vertexai-fc-pr (retrieval-app-testing)"
 # Set team access
 permissionRules:
   - team: senseai-eco

.github/workflows/lint.yml

@@ -34,7 +34,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        dir: [retrieval_service, llm_demo]
+        dir: [retrieval_service]
       fail-fast: false
     permissions:
       contents: read

.github/workflows/lint_fallback.yml

@@ -25,7 +25,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        dir: [retrieval_service, llm_demo]
+        dir: [retrieval_service]
     permissions:
       contents: none
 

llm_demo/Dockerfile renamed to Dockerfile

@@ -27,7 +27,7 @@
 from markdown import markdown
 from starlette.middleware.sessions import SessionMiddleware
 
-from orchestrator import createOrchestrator
+from orchestrator import Orchestrator
 
 routes = APIRouter()
 templates = Jinja2Templates(directory="templates")
@@ -237,16 +237,13 @@ def clear_user_info(session: dict[str, Any]):
 
 
 def init_app(
-    orchestration_type: Optional[str],
     client_id: Optional[str],
     middleware_secret: Optional[str],
 ) -> FastAPI:
     # FastAPI setup
-    if orchestration_type is None:
-        raise HTTPException(status_code=500, detail="Orchestrator not found")
     app = FastAPI(lifespan=lifespan)
     app.state.client_id = client_id
-    app.state.orchestrator = createOrchestrator(orchestration_type)
+    app.state.orchestrator = Orchestrator()
     app.include_router(routes)
     app.mount("/static", StaticFiles(directory="static"), name="static")
     app.add_middleware(SessionMiddleware, secret_key=middleware_secret)
@@ -256,12 +253,9 @@ def init_app(
 if __name__ == "__main__":
     PORT = int(os.getenv("PORT", default=8081))
     HOST = os.getenv("HOST", default="0.0.0.0")
-    ORCHESTRATION_TYPE = os.getenv("ORCHESTRATION_TYPE", default="langchain-tools")
     CLIENT_ID = os.getenv("CLIENT_ID")
     MIDDLEWARE_SECRET = os.getenv("MIDDLEWARE_SECRET", default="this is a secret")
-    app = init_app(
-        ORCHESTRATION_TYPE, client_id=CLIENT_ID, middleware_secret=MIDDLEWARE_SECRET
-    )
+    app = init_app(client_id=CLIENT_ID, middleware_secret=MIDDLEWARE_SECRET)
     if app is None:
         raise TypeError("app not instantiated")
     uvicorn.run(app, host=HOST, port=PORT)

llm_demo/app.py renamed to app.py

@@ -14,14 +14,11 @@
 steps:
   - id: Install dependencies
     name: python:3.11
-    dir: llm_demo 
     script: pip install -r requirements.txt -r requirements-test.txt --user
 
   - id: "Run evaluation service"
     name: python:3.11
-    dir: llm_demo
     env: # Set env var expected by tests
-      - "ORCHESTRATION_TYPE=${_ORCHESTRATION_TYPE}"
       - "RETRIEVAL_EXPERIMENT_NAME=${_RETRIEVAL_EXPERIMENT_NAME}"
       - "RESPONSE_EXPERIMENT_NAME=${_RESPONSE_EXPERIMENT_NAME}"
     secretEnv:
@@ -37,7 +34,6 @@ options:
   dynamic_substitutions: true
 
 substitutions:
-  _ORCHESTRATION_TYPE: "langchain-tools"
   _RETRIEVAL_EXPERIMENT_NAME: "retrieval-phase-eval-${_PR_NUMBER}"
   _RESPONSE_EXPERIMENT_NAME:  "response-phase-eval-${_PR_NUMBER}"
 

llm_demo/app_test.py renamed to app_test.py

@@ -17,22 +17,20 @@
 from typing import Dict, List
 
 import pandas as pd
-from pydantic import BaseModel, Field
 from vertexai.evaluation import EvalTask
 from vertexai.evaluation import _base as evaluation_base
 
-from orchestrator import BaseOrchestrator
+from orchestrator import Orchestrator
 
 from .eval_golden import EvalData, ToolCall
 from .metrics import response_phase_metrics, retrieval_phase_metrics
 
 
 async def run_llm_for_eval(
-    eval_list: List[EvalData], orc: BaseOrchestrator, session: Dict, session_id: str
+    eval_list: List[EvalData], orc: Orchestrator, session: Dict, session_id: str
 ) -> List[EvalData]:
     """
     Generate llm_tool_calls and llm_output for golden dataset query.
-    This function is only compatible with the langchain-tools orchestration.
     """
     agent = orc.get_user_session(session_id)
     for eval_data in eval_list: