add basic (test only) openai integration

Author: MoSchmidt

backend/.env.example

@@ -11,3 +11,5 @@ DATABASE_URL=postgresql+psycopg2://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTG
 # Authentication
 JWT_SECRET_KEY=3e3cc0b2f8b54b3d6aa597faccb2f9c41b19b1c24befe29661bafbd42e3afe3e
 
+# OpenAI
+OPENAI_API_KEY=my_secret_key

backend/app/core/config.py

@@ -24,6 +24,9 @@ class Settings(BaseSettings):
     ACCESS_TOKEN_EXPIRE_MINUTES: int = 15
     REFRESH_TOKEN_EXPIRE_DAYS: int = 7
 
+    # --- OpenAI ---
+    OPENAI_API_KEY: str
+
     model_config = SettingsConfigDict(
         env_file=os.getenv("ENV_FILE", "dev.env"),
         extra="ignore",

backend/app/core/deps.py

@@ -0,0 +1,9 @@
+from app.llm.openai.provider import OpenAIProvider
+
+_openai_provider = None
+
+def get_openai_provider() -> OpenAIProvider:
+    global _openai_provider
+    if _openai_provider is None:
+        _openai_provider = OpenAIProvider()
+    return _openai_provider

backend/app/llm/__init__.py

@@ -0,0 +1,14 @@
+KEYWORD_PROMPT="""
+You are an expert in academic information retrieval. Extract 7–15 single-word academic keywords suitable for searching scientific databases (arXiv, IEEE, ACL).
+
+Rules:
+
+- Single words only.
+- Academic, domain-specific terms only (tasks, algorithms, architectures, datasets, phenomena).
+- Exclude broad or high-frequency terms (e.g., learning, inference, neural, optimization, classification, system, model, method).
+- No generic research words (e.g., study, analysis, results, approach).
+- No overlapping or substring-related keywords.
+- Each keyword must significantly narrow the search space — prefer specific technical terminology used mainly within a subfield.
+
+Output only a JSON list: ["keyword1", ...].
+"""

backend/app/llm/openai/__init__.py

@@ -0,0 +1,31 @@
+from openai import OpenAI
+
+from app.core.config import settings
+from app.llm.openai.prompts import KEYWORD_PROMPT
+
+
+class OpenAIProvider:
+    """Wrapper around the OpenAI client."""
+
+    _model: str
+
+    def __init__(self) -> None:
+        self.client = OpenAI(api_key=settings.OPENAI_API_KEY)
+        self._model = "gpt-5-nano-2025-08-07"
+
+    def extract_keywords(self, user_text: str) -> str:
+        response = self.client.responses.create(
+            model=self._model,
+            reasoning={"effort": "low"},
+            input=[
+                {
+                    "role": "developer",
+                    "content":  KEYWORD_PROMPT,
+                },
+                {
+                    "role": "user",
+                    "content": user_text,
+                },
+            ],
+        )
+        return response.output_text

backend/app/llm/openai/prompts.py

@@ -7,7 +7,7 @@
 
 from app.core.config import settings
 from app.core.database import init_db
-from app.routes import auth_routes, user_routes
+from app.routes import auth_routes, user_routes, search_routes
 
 # ---------------------------------------------------------
 # Configure Logging
@@ -68,3 +68,4 @@ async def lifespan(_app: FastAPI) -> AsyncGenerator[None, Any]:
 # ---------------------------------------------------------
 app.include_router(user_routes.router)
 app.include_router(auth_routes.router)
+app.include_router(search_routes.router)

backend/app/llm/openai/provider.py

@@ -0,0 +1,13 @@
+from fastapi import APIRouter, status, Depends
+
+from app.core.deps import get_openai_provider
+from app.llm.openai.provider import OpenAIProvider
+from app.schemas.search_dto import SearchRequest
+
+router = APIRouter(prefix="/search", tags=["Search"])
+
+# TODO remove - testing purposes only
+@router.post("/", response_model=str, status_code=status.HTTP_200_OK, summary="Authenticate a user and return JWT tokens")
+def search(search_request: SearchRequest, provider: OpenAIProvider = Depends(get_openai_provider)) -> str:
+    return provider.extract_keywords(search_request.search_text)
+

backend/app/main.py

@@ -0,0 +1,5 @@
+from pydantic import BaseModel
+
+
+class SearchRequest(BaseModel):
+    search_text: str