Implement basic_agent with llama-index

.gitignore

@@ -0,0 +1,2 @@
+.venv
+tool_tests.ipynb

app.py

@@ -4,12 +4,20 @@ import requests
 import inspect
 import pandas as pd
 
+from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+from llama_index.core import Settings
+
 from basic_agent import BasicAgent
 
 # (Keep Constants as is)
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
+Settings.llm = None # disable LLM for Index Retrieval
+Settings.chunk_size = 512 # Smaller chunk size for retrieval
+
+Settings.embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-small-en-v1.5")
+
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
     Fetches all questions, runs the BasicAgent on them, submits all answers,

basic_agent.py

@@ -3,23 +3,45 @@ from llama_index.core.workflow import Context
 from llama_index.core.tools import FunctionTool
 from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
 from llama_index.tools.duckduckgo import DuckDuckGoSearchToolSpec
+from llama_index.tools.wikipedia import WikipediaToolSpec
+from llama_index.tools.tool_spec.load_and_search.base import LoadAndSearchToolSpec
+from llama_index.readers.web import SimpleWebPageReader
+from llama_index.core.tools.ondemand_loader_tool import OnDemandLoaderTool
 
 class BasicAgent:
     def __init__(self):
-      llm = HuggingFaceInferenceAPI(model_name="Qwen/Qwen2.5-Coder-32B-Instruct")
+        llm = HuggingFaceInferenceAPI(model_name="Qwen/Qwen2.5-Coder-32B-Instruct")
 
-      # Initialize tools
-      tool_spec = DuckDuckGoSearchToolSpec()
-      search_tool = FunctionTool.from_defaults(tool_spec.duckduckgo_full_search)
-      
-      # Create Alfred with all the tools
-      self.agent = AgentWorkflow.from_tools_or_functions(
-          [search_tool],
-          llm=llm
-      )
+        # Initialize tools
+        tool_spec = DuckDuckGoSearchToolSpec()
+        search_tool = FunctionTool.from_defaults(tool_spec.duckduckgo_full_search)
 
-      # self.ctx = Context(self.agent)
+        wiki_spec = WikipediaToolSpec()
+        wiki_search_tool = wiki_spec.to_tool_list()[1]
+
+        # Convert into a LoadAndSearchToolSpec because the wikipedia search tool returns
+        # entire Wikipedia pages and this can pollute the context window of the LLM
+        wiki_spec = WikipediaToolSpec()
+        wiki_search_tool = wiki_spec.to_tool_list()[1]
+
+        # Convert into a LoadAndSearchToolSpec because the wikipedia search tool returns
+        # entire Wikipedia pages and this can pollute the context window of the LLM
+        wiki_search_tool_las = LoadAndSearchToolSpec.from_defaults(wiki_search_tool).to_tool_list()
+
+        webpage_tool = OnDemandLoaderTool.from_defaults(
+            SimpleWebPageReader(html_to_text=True),
+            name="Webpage search tool",
+            description="A tool for loading the content of a webpage and querying it for information",
+        )
+
+        # Create Alfred with all the tools
+        self.agent = AgentWorkflow.from_tools_or_functions(
+            [search_tool, wiki_search_tool_las, webpage_tool],
+            llm=llm
+        )
+
+        # self.ctx = Context(self.agent)
 
     async def __call__(self, question: str) -> str:
-      response = await self.agent.run(user_msg=question) # ctx=self.ctx)
-      return response.response.content
+        response = await self.agent.run(user_msg=question) # ctx=self.ctx)
+        return response.response.content

requirements.txt

@@ -2,4 +2,8 @@ gradio
 requests
 llama-index
 llama-index-llms-huggingface-api
-llama_index-tools-duckduckgo
+llama_index-tools-duckduckgo
+llama_index-tools-wikipedia
+llama-index-indices-managed-bge-m3
+llama-index-embeddings-huggingface
+llama-index-readers-web