Return solutions which may have been hallucinated

README.md

@@ -1,10 +1,10 @@
 ---
-title: Reqroup
+title: Reqxtract-API
 emoji: 🤖
 colorFrom: purple
 colorTo: blue
 sdk: docker
 app_port: 8000
 pinned: false
-short_description: Categorize service requirements into groups (using AI)
+short_description: API for Reqxtract.
 ---

app.py

@@ -43,9 +43,12 @@ llm_router = Router(model_list=[
 prompt_env = Environment(loader=FileSystemLoader(
     'prompts'), enable_async=True, undefined=StrictUndefined)
 
-api = FastAPI(docs_url="/", title="Reqxtract-API", description=open("docs/docs.md").read())
+api = FastAPI(docs_url="/", title="Reqxtract-API",
+              description=open("docs/docs.md").read())
+
 # requirements routes
 requirements_router = APIRouter(prefix="/reqs", tags=["requirements"])
+
 # solution routes
 solution_router = APIRouter(prefix="/solution", tags=["solution"])
 
@@ -144,9 +147,13 @@ async def search_solutions(params: ReqGroupingResponse) -> SolutionSearchRespons
 
         # ======================== build the final solution object ================================
 
-        # extract the source metadata from the search items
-        sources_metadata = [{"name": a["web"]["title"], "url": a["web"]["uri"]}
-                            for a in req_completion["vertex_ai_grounding_metadata"][0]['groundingChunks']]
+        sources_metadata = []
+        # extract the source metadata from the search items, if gemini actually called the tools to search .... and didn't hallucinated
+        try:
+            sources_metadata.extend([{"name": a["web"]["title"], "url": a["web"]["uri"]}
+                                     for a in req_completion["vertex_ai_grounding_metadata"][0]['groundingChunks']])
+        except KeyError as ke:
+            pass
 
         final_sol = SolutionModel(
             Context="",