Updated gemini models (#1528)

* Updated gemini models

* added tooltip

* updated packages

* gds related changes

---------

Co-authored-by: Prakriti Solankey <156313631+prakriti-solankey@users.noreply.github.com>

Author: kaustubh-darekar

README.md

@@ -114,7 +114,7 @@ Run the application using the default `docker-compose` configuration.
    By default, only OpenAI and Diffbot are enabled. Gemini requires additional GCP configurations.  
    Use the `VITE_LLM_MODELS_PROD` variable to configure the models you need. Example:
    ```bash
-   VITE_LLM_MODELS_PROD="gemini_2.5_flash,openai_gpt_5_mini,diffbot,anthropic_claude_4.5_haiku"
+   VITE_LLM_MODELS_PROD="gemini_flash_latest,openai_gpt_5_mini,diffbot,anthropic_claude_4.5_haiku"
    ```
 
 2. **Input Sources:**  
@@ -267,12 +267,12 @@ gcloud run deploy dev-backend \
 | VITE_REACT_APP_SOURCES | Mandatory          | local,youtube,wiki,s3 | List of input sources that will be available                                 |
 | VITE_CHAT_MODES        | Mandatory          | vector,graph+vector,graph,hybrid | Chat modes available for Q&A                                |
 | VITE_ENV               | Mandatory          | DEV or PROD   | Environment variable for the app                                                                 |
-| VITE_LLM_MODELS        | Optional           | openai_gpt_5_mini,gemini_2.5_flash,anthropic_claude_4.5_haiku | Supported models for the application |
+| VITE_LLM_MODELS        | Optional           | openai_gpt_5_mini,gemini_flash_latest,anthropic_claude_4.5_haiku | Supported models for the application |
 | VITE_BACKEND_API_URL   | Optional           | [localhost][backend-url] | URL for backend API                                        |
 | VITE_TIME_PER_PAGE     | Optional           | 50            | Time per page for processing                                                                     |
 | VITE_CHUNK_SIZE        | Optional           | 5242880       | Size of each chunk of file for upload                                                            |
 | VITE_GOOGLE_CLIENT_ID  | Optional           |               | Client ID for Google authentication                                                              |
-| VITE_LLM_MODELS_PROD   | Optional           | openai_gpt_5_mini,gemini_2.5_flash,anthropic_claude_4.5_haiku | To distinguish models based on environment (PROD or DEV) |
+| VITE_LLM_MODELS_PROD   | Optional           | openai_gpt_5_mini,gemini_flash_latest,anthropic_claude_4.5_haiku | To distinguish models based on environment (PROD or DEV) |
 | VITE_AUTH0_CLIENT_ID   | Mandatory if you are enabling Authentication otherwise it is optional |  | Okta OAuth Client ID for authentication |
 | VITE_AUTH0_DOMAIN      | Mandatory if you are enabling Authentication otherwise it is optional |  | Okta OAuth Client Domain                                  |
 | VITE_SKIP_AUTH         | Optional           | true          | Flag to skip authentication                                                                      |

backend/example.env

@@ -29,8 +29,8 @@ LLM_MODEL_CONFIG_openai_gpt_5.2="gpt-5.2,openai-key"
 LLM_MODEL_CONFIG_openai_gpt_5_mini="gpt-5-mini,openai-key"
 LLM_MODEL_CONFIG_openai_gpt_4.1="gpt-4.1,openai-key"
 LLM_MODEL_CONFIG_openai_gpt_4.1_mini="gpt-4.1-mini,openai-key"
-LLM_MODEL_CONFIG_gemini_2.5_flash="gemini-2.5-flash"
-LLM_MODEL_CONFIG_gemini_2.5_pro="gemini-2.5-pro"
+LLM_MODEL_CONFIG_gemini_flash_latest="gemini-flash-latest"
+LLM_MODEL_CONFIG_gemini_pro_latest="gemini-pro-latest"
 LLM_MODEL_CONFIG_diffbot="diffbot,diffbot_api_key"
 LLM_MODEL_CONFIG_groq_llama3.1_8b="llama-3.1-8b-instant,base_url,groq_api_key"
 LLM_MODEL_CONFIG_anthropic_claude_4.5_sonnet="claude-sonnet-4-5-20250929,anthropic_api_key"

backend/requirements.txt

@@ -1,64 +1,64 @@
-accelerate==1.12.0
+accelerate==1.13.0
 asyncio==4.0.0
-boto3==1.40.23
-botocore==1.40.23
-certifi==2025.8.3
-fastapi==0.116.1
+boto3==1.42.66
+botocore==1.42.66
+certifi==2026.2.25
+fastapi==0.135.1
 fastapi-health==0.4.0
-fireworks-ai==0.15.12
-google-api-core==2.25.1
-google-auth==2.40.3
-google_auth_oauthlib==1.2.2
-google-cloud-core==2.4.3
+fireworks-ai==0.19.20
+google-api-core==2.30.0
+google-auth==2.49.0
+google_auth_oauthlib==1.3.0
+google-cloud-core==2.5.0
 json-repair==0.44.1
 pip-install==1.3.5
-langchain==1.1.2
-langchain-aws==1.1.0
-langchain-anthropic==1.2.0
+langchain==1.2.12
+langchain-aws==1.4.0
+langchain-anthropic==1.3.4
 langchain-fireworks==1.1.0
 langchain-community==0.4.1
-langchain-core==1.2.5
-langchain-experimental==0.4.0
-langchain-google-vertexai==3.1.1
-langchain-groq==1.1.0
-langchain-openai==1.1.0
-langchain-text-splitters==1.0.0
-langchain-huggingface==1.1.0
+langchain-core==1.2.18
+langchain-experimental==0.4.1
+langchain_google_genai==4.2.1
+langchain-groq==1.1.2
+langchain-openai==1.1.11
+langchain-text-splitters==1.1.1
+langchain-huggingface==1.2.1
 langchain-classic==1.0.0
 langdetect==1.0.9
 langsmith==0.4.55
 langserve==0.3.3
 neo4j-rust-ext==5.28.2.1
-nltk==3.9.1
-openai==2.9.0
-psutil==7.0.0
+nltk==3.9.3
+openai==2.26.0
+psutil==7.2.2
 pydantic==2.12.5
 python-dotenv==1.1.1
 python-magic==0.4.27
 PyPDF2==3.0.1
-PyMuPDF==1.26.4
+PyMuPDF==1.27.2
 starlette==0.47.3
 sse-starlette==3.0.2
 starlette-session==0.4.3
-tqdm==4.67.1
-unstructured[all-docs]==0.18.14
-unstructured-client==0.42.3
-unstructured-inference==1.0.5
-urllib3==2.6.2
-uvicorn==0.35.0
-gunicorn==23.0.0
+tqdm==4.67.3
+unstructured[all-docs]==0.21.5
+unstructured-client==0.42.10
+unstructured-inference==1.2.0
+urllib3==2.2.2
+uvicorn==0.41.0
+gunicorn==25.1.0
 wikipedia==1.4.0
 wrapt==1.17.3
 yarl==1.20.1
-youtube-transcript-api==1.2.2
+youtube-transcript-api==1.2.4
 zipp==3.23.0
 sentence-transformers==5.1.0
 google-cloud-logging==3.12.1
-pypandoc==1.15
-graphdatascience==1.18a1
+pypandoc==1.16.2
+graphdatascience==1.20
 Secweb==1.25.2
-ragas==0.4.0
+ragas==0.4.3
 rouge_score==0.1.2
-langchain-neo4j==0.6.0
-pypandoc-binary==1.15
+langchain-neo4j==0.7.0
+pypandoc-binary==1.16.2
 chardet==5.2.0

backend/score.py

@@ -1019,12 +1019,14 @@ async def calculate_additional_metrics(question: str = Form(),
                                         reference: str = Form(),
                                         model: str = Form(),
                                         mode: str = Form(),
+                                        embedding_provider: str = Form(None),
+                                        embedding_model: str = Form(None)
 ):
    try:
        context_list = [str(item).strip() for item in json.loads(context)] if context else []
        answer_list = [str(item).strip() for item in json.loads(answer)] if answer else []
        mode_list = [str(item).strip() for item in json.loads(mode)] if mode else []
-       result = await get_additional_metrics(question, context_list,answer_list, reference, model)
+       result = await get_additional_metrics(question, context_list,answer_list, reference, model, embedding_provider, embedding_model)
        if result is None or "error" in result:
            return create_api_response(
                'Failed',

backend/src/QA_integration.py

@@ -8,9 +8,7 @@
 from typing import Any
 from dotenv import load_dotenv
 
-from langchain_neo4j import Neo4jVector
-from langchain_neo4j import Neo4jChatMessageHistory
-from langchain_neo4j import GraphCypherQAChain
+from langchain_neo4j import Neo4jVector, Neo4jChatMessageHistory, GraphCypherQAChain
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 from langchain_core.output_parsers import StrOutputParser
 from langchain_core.runnables import RunnableBranch
@@ -20,10 +18,10 @@
 from langchain_core.messages import HumanMessage, AIMessage
 from langchain_community.chat_message_histories import ChatMessageHistory 
 from langchain_core.callbacks import StdOutCallbackHandler, BaseCallbackHandler
-from src.shared.llm_graph_builder_exception import LLMGraphBuilderException
+# from src.shared.llm_graph_builder_exception import LLMGraphBuilderException
 # LangChain chat models
 from langchain_openai import ChatOpenAI, AzureChatOpenAI
-from langchain_google_vertexai import ChatVertexAI
+from langchain_google_genai import ChatGoogleGenerativeAI
 from langchain_groq import ChatGroq
 from langchain_anthropic import ChatAnthropic
 from langchain_fireworks import ChatFireworks
@@ -75,8 +73,12 @@ def get_total_tokens(ai_response, llm):
         if isinstance(llm, (ChatOpenAI, AzureChatOpenAI, ChatFireworks, ChatGroq)):
             total_tokens = ai_response.response_metadata.get('token_usage', {}).get('total_tokens', 0)
 
-        elif isinstance(llm, ChatVertexAI):
-            total_tokens = ai_response.response_metadata.get('usage_metadata', {}).get('prompt_token_count', 0)
+        elif isinstance(llm, ChatGoogleGenerativeAI):
+            if hasattr(ai_response, 'usage_metadata') and ai_response.usage_metadata:
+                total_tokens = ai_response.usage_metadata.get('total_tokens', 0)
+            else:
+                usage = ai_response.response_metadata.get('token_usage', {}) or ai_response.response_metadata.get('usage_metadata', {})
+                total_tokens = usage.get('total_tokens', 0)
 
         elif isinstance(llm, ChatBedrock):
             total_tokens = ai_response.response_metadata.get('usage', {}).get('total_tokens', 0)
@@ -224,6 +226,13 @@ def format_documents(documents, model,chat_mode_settings):
 
     return "\n\n".join(formatted_docs), sources,entities,global_communities
 
+def get_clean_text(msg):
+    if isinstance(msg.content, str):
+        return msg.content
+    return msg.additional_kwargs.get("text", "") or "".join(
+        [p.get("text", "") for p in msg.content if isinstance(p, dict)]
+    )
+    
 def process_documents(docs, question, messages, llm, model,chat_mode_settings):
     start_time = time.time()
 
@@ -256,7 +265,7 @@ def process_documents(docs, question, messages, llm, model,chat_mode_settings):
         result["nodedetails"] = node_details
         result["entities"] = entities
 
-        content = ai_response.content
+        content = get_clean_text(ai_response)
         total_tokens = get_total_tokens(ai_response, llm)
 
         predict_time = time.time() - start_time
@@ -508,33 +517,41 @@ def summarize_and_log(history, stored_messages, llm):
     try:
         start_time = time.time()
 
-        summarization_prompt = ChatPromptTemplate.from_messages(
-            [
-                MessagesPlaceholder(variable_name="chat_history"),
-                (
-                    "human",
-                    "Summarize the above chat messages into a concise message, focusing on key points and relevant details that could be useful for future conversations. Exclude all introductions and extraneous information."
-                ),
-            ]
-        )
+        summarization_prompt = ChatPromptTemplate.from_messages([
+            MessagesPlaceholder(variable_name="chat_history"),
+            ("human", "Summarize the above chat messages into a concise message..."),
+        ])
+
         summarization_chain = summarization_prompt | llm
 
-        summary_message = summarization_chain.invoke({"chat_history": stored_messages})
+        raw_summary = summarization_chain.invoke({"chat_history": stored_messages})
+
+        if hasattr(raw_summary, "content"):
+            content = raw_summary.content
+            if isinstance(content, list):
+                summary_text = "".join([
+                    block.get("text", "") if isinstance(block, dict) else str(block) 
+                    for block in content
+                ])
+            else:
+                summary_text = str(content)
+        else:
+            summary_text = str(raw_summary)
+
+        summary_message_for_db = AIMessage(content=summary_text)
 
         with threading.Lock():
             history.clear()
             history.add_user_message("Our current conversation summary till now")
-            history.add_message(summary_message)
-
-        history_summarized_time = time.time() - start_time
-        logging.info(f"Chat History summarized in {history_summarized_time:.2f} seconds")
+            history.add_message(summary_message_for_db)
 
+        logging.info(f"Chat History summarized in {time.time() - start_time:.2f} seconds")
         return True
 
     except Exception as e:
         logging.error(f"An error occurred while summarizing messages: {e}", exc_info=True)
-        return False 
-    
+        return False
+      
 def create_graph_chain(model, graph):
     try:
         logging.info(f"Graph QA Chain using LLM model: {model}")

backend/src/graphDB_dataAccess.py

@@ -199,22 +199,27 @@ def check_account_access(self, database):
             return False
 
     def check_gds_version(self):
-        try:
-            gds_procedure_count = """
-            SHOW FUNCTIONS YIELD name WHERE name STARTS WITH 'gds.version' RETURN COUNT(*) AS totalGdsProcedures
-            """
-            result = self.graph.query(gds_procedure_count,session_params={"database":self.graph._database})
-            total_gds_procedures = result[0]['totalGdsProcedures'] if result else 0
+            try:
+                gds_check_query = "RETURN gds.version() AS gdsVersion"
+                result = self.graph.query(gds_check_query, session_params={"database": self.graph._database})
+                
+                version = result[0]['gdsVersion'] if result else None
 
-            if total_gds_procedures > 0:
-                logging.info("GDS is available in the database.")
-                return True
-            else:
-                logging.info("GDS is not available in the database.")
-                return False
-        except Exception as e:
-            logging.error(f"An error occurred while checking GDS version: {e}")
-            return False
+                if version:
+                    logging.info(f"GDS compute engine is available. Version: {version}")
+                    return True
+                else:
+                    logging.info("GDS check returned empty.")
+                    return False
+                    
+            except Exception as e:
+                error_msg = str(e)
+                if "versionless" in error_msg or "Unknown function" in error_msg:
+                    logging.info("GDS compute engine is not provisioned on this database instance.")
+                    return False
+                else:
+                    logging.error(f"An unexpected error occurred while verifying GDS: {e}")
+                    return False
 
     def connection_check_and_get_vector_dimensions(self, database, email, uri):
         """

backend/src/llm.py

@@ -2,9 +2,8 @@
 from langchain_core.documents import Document
 import os
 from langchain_openai import ChatOpenAI, AzureChatOpenAI
-from langchain_google_vertexai import ChatVertexAI
+from langchain_google_genai import ChatGoogleGenerativeAI
 from langchain_groq import ChatGroq
-from langchain_google_vertexai import HarmBlockThreshold, HarmCategory
 from langchain_experimental.graph_transformers.diffbot import DiffbotGraphTransformer
 from langchain_experimental.graph_transformers import LLMGraphTransformer
 from langchain_experimental.graph_transformers.llm import _Graph
@@ -39,18 +38,19 @@ def get_llm(model: str):
         if "GEMINI" in model:
             model_name = env_value
             credentials, project_id = google.auth.default()
-            llm = ChatVertexAI(
-                model_name=model_name,
+            llm = ChatGoogleGenerativeAI(
+                model=model_name,
+                vertexai=True,
                 credentials=credentials,
                 project=project_id,
                 temperature=0,
                 callbacks=callback_manager,
                 safety_settings={
-                    HarmCategory.HARM_CATEGORY_UNSPECIFIED: HarmBlockThreshold.BLOCK_NONE,
-                    HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_NONE,
-                    HarmCategory.HARM_CATEGORY_HATE_SPEECH: HarmBlockThreshold.BLOCK_NONE,
-                    HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_NONE,
-                    HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: HarmBlockThreshold.BLOCK_NONE,
+                    "HARM_CATEGORY_UNSPECIFIED": "BLOCK_NONE",
+                    "HARM_CATEGORY_DANGEROUS_CONTENT": "BLOCK_NONE",
+                    "HARM_CATEGORY_HATE_SPEECH": "BLOCK_NONE",
+                    "HARM_CATEGORY_HARASSMENT": "BLOCK_NONE",
+                    "HARM_CATEGORY_SEXUALLY_EXPLICIT": "BLOCK_NONE",
                 },
 
             )

backend/src/shared/common_fn.py

@@ -10,7 +10,7 @@
 import logging
 from urllib.parse import urlparse,parse_qs
 from src.shared.llm_graph_builder_exception import LLMGraphBuilderException
-from langchain_google_vertexai import VertexAIEmbeddings
+from langchain_google_genai import GoogleGenerativeAIEmbeddings
 from langchain_openai import OpenAIEmbeddings
 from langchain_neo4j import Neo4jGraph
 from neo4j.exceptions import TransientError
@@ -200,7 +200,7 @@ def load_embedding_model(embedding_provider: str, embedding_model_name: str):
     if provider == "openai":
         embeddings = OpenAIEmbeddings(model=model)
     elif provider == "gemini":
-        embeddings = VertexAIEmbeddings(model=model)
+        embeddings = GoogleGenerativeAIEmbeddings(model=model, vertexai= True)
     elif provider == "titan":
         embeddings = _get_bedrock_embeddings(model)
     elif provider == "sentence-transformer":

backend/src/shared/constants.py

@@ -247,7 +247,7 @@
 CHAT_EMBEDDING_FILTER_SCORE_THRESHOLD = 0.10
 
 CHAT_TOKEN_CUT_OFF = {
-     ("gemini_2.5_pro", "gemini_2.5_flash","groq-llama3",'groq_llama3_70b','anthropic_claude_4_5_sonnet','anthropic_claude_4') : 4, 
+     ("gemini_pro_latest", "gemini_flash_latest","groq-llama3",'groq_llama3_70b','anthropic_claude_4_5_sonnet','anthropic_claude_4') : 4, 
      ("openai_gpt_5.2","diffbot","openai_gpt_5_mini") : 28,
      ("ollama_llama3") : 2  
 }  

backend/src/shared/schema_extraction.py

@@ -1,5 +1,5 @@
 from typing import List
-from pydantic.v1 import BaseModel, Field
+from pydantic import BaseModel, Field
 from src.llm import get_llm
 from langchain_core.prompts import ChatPromptTemplate
 import logging