Fix fallback for keyword-based search

docker-compose.yaml

@@ -17,13 +17,13 @@ services:
       - "5173:5173"
     environment:
       REST_API_URL: http://localhost:8000
-    volumes:
-      - ./frontend:/app
+    # volumes:
+    #   - ./frontend:/app
 
   rest:
     image: ghcr.io/catalystneuro/dandiset-search-service:latest
     container_name: dandi-search-rest
-    command: ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000", "--workers", "4"]
+    command: ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000", "--workers", "4", "--reload"]
     ports:
       - "8000:8000"
     environment:
@@ -34,3 +34,5 @@ services:
       QDRANT_API_KEY: ${QDRANT_API_KEY}
       OPENAI_API_KEY: ${OPENAI_API_KEY}
       DANDI_API_KEY: ${DANDI_API_KEY}
+    volumes:
+      - ./rest:/app

rest/clients/openai.py

@@ -92,15 +92,24 @@ def keywords_extraction(self, user_input: str, model: str = "gpt-3.5-turbo"):
         )
         chain = create_extraction_chain(schema, llm)
         keywords_extracted = list(chain.run(user_input))
+
+        # Temporary fallback to extracting nouns (if no schema-related keywords found)
         if any(isinstance(item, str) for item in keywords_extracted):
-            keywords_extracted = [{key: ""} for key in schema["properties"]]
+            # import nltk 
+            # nltk.download(['punkt', 'averaged_perceptron_tagger'])
+            # words = nltk.word_tokenize(user_input)
+            # pos_tags = nltk.pos_tag(words)
+            # keywords_extracted = [noun for noun, tag in pos_tags if tag.startswith("N")]
+            # return keywords_extracted
+            return list(set(user_input.split()))
+
         return self.prepare_keywords_for_semantic_search(keywords_extracted)
 
 
     def prepare_keywords_for_semantic_search(self, keywords_list: list) -> list:
         keywords_set = set()
         for obj in keywords_list:
-            for k, v in obj.items():
+            for _, v in obj.items():
                 if v:
                     keywords_set.add(v.lower())
         return list(keywords_set)