File size: 13,938 Bytes
a3e120b
 
 
 
bf3da7a
a3e120b
 
e5bb694
bd93d23
e5bb694
6a1d4f7
 
 
 
e5bb694
 
a3e120b
6a1d4f7
bf3da7a
6a1d4f7
 
 
e5bb694
6a1d4f7
e5bb694
6a1d4f7
e5bb694
d5ef142
e5bb694
a3e120b
e5bb694
bf3da7a
6a1d4f7
 
 
e5bb694
6a1d4f7
 
1376719
a3e120b
 
e5bb694
bf3da7a
d5ef142
bf3da7a
 
d5ef142
 
 
 
e5bb694
bf3da7a
e5bb694
d5ef142
 
 
 
 
bf3da7a
 
 
 
 
 
 
 
 
 
 
 
 
d5ef142
bf3da7a
 
 
 
 
 
1376719
bf3da7a
 
 
1376719
bf3da7a
 
 
 
1376719
bf3da7a
 
 
bd93d23
bf3da7a
 
 
 
 
 
bd93d23
bf3da7a
 
 
 
 
 
 
 
1376719
bf3da7a
 
 
 
bd93d23
bf3da7a
bd93d23
bf3da7a
 
 
 
 
 
bd93d23
bf3da7a
 
 
 
1376719
bf3da7a
 
 
 
 
 
 
bd93d23
bf3da7a
a3e120b
bf3da7a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a3e120b
bf3da7a
 
 
 
a3e120b
bf3da7a
 
 
 
 
 
 
 
 
 
 
6a1d4f7
 
bf3da7a
6a1d4f7
bf3da7a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1376719
bf3da7a
 
 
 
 
 
 
 
 
0e1e580
bf3da7a
 
 
 
 
1376719
6a1d4f7
 
bd93d23
e5bb694
d5ef142
 
 
 
 
 
bf3da7a
 
 
 
 
d4a60c1
bf3da7a
 
 
 
 
 
 
1376719
bf3da7a
d5ef142
 
bf3da7a
 
 
d5ef142
bd93d23
bf3da7a
 
 
 
 
 
 
 
1376719
bf3da7a
 
 
1376719
bf3da7a
bd93d23
1376719
bf3da7a
1376719
d5ef142
 
bf3da7a
d5ef142
bf3da7a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1376719
bf3da7a
 
 
 
 
d5ef142
 
bf3da7a
e5bb694
bf3da7a
e5bb694
bf3da7a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e5bb694
bf3da7a
e5bb694
 
bf3da7a
e5bb694
bf3da7a
 
 
 
 
 
e5bb694
 
 
bf3da7a
bd93d23
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
"""
agent.py

This file defines the core logic for a sophisticated AI agent using LangGraph.
This version uses Groq's vision-capable models and includes proper reasoning steps.
"""

# ----------------------------------------------------------
# Section 0: Imports and Configuration
# ----------------------------------------------------------
import json
import os
import pickle
import re
import subprocess
import textwrap
import functools
from pathlib import Path
from typing import Dict, Any

import requests
from cachetools import TTLCache

from langchain.schema import Document
from langchain.tools.retriever import create_retriever_tool
from langchain_community.vectorstores import FAISS
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.document_loaders import WikipediaLoader, ArxivLoader
from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
from langchain_core.tools import Tool, tool
from langchain_groq import ChatGroq
from langchain_huggingface import HuggingFaceEmbeddings
from langgraph.graph import START, StateGraph, MessagesState
from langgraph.prebuilt import ToolNode, tools_condition

from dotenv import load_dotenv
load_dotenv()

# --- Configuration and Caching ---
JSONL_PATH, FAISS_CACHE, EMBED_MODEL = Path("metadata.jsonl"), Path("faiss_index.pkl"), "sentence-transformers/all-mpnet-base-v2"
RETRIEVER_K, CACHE_TTL = 5, 600
API_CACHE = TTLCache(maxsize=256, ttl=CACHE_TTL)

def cached_get(key: str, fetch_fn):
    if key in API_CACHE: 
        return API_CACHE[key]
    val = fetch_fn()
    API_CACHE[key] = val
    return val

# ----------------------------------------------------------
# Section 2: Tool Functions
# ----------------------------------------------------------
@tool
def python_repl(code: str) -> str:
    """Executes a string of Python code and returns the stdout/stderr."""
    code = textwrap.dedent(code).strip()
    try:
        result = subprocess.run(
            ["python", "-c", code], 
            capture_output=True, 
            text=True, 
            timeout=10, 
            check=False
        )
        if result.returncode == 0: 
            return f"Execution successful.\nSTDOUT:\n```\n{result.stdout}\n```"
        else: 
            return f"Execution failed.\nSTDOUT:\n```\n{result.stdout}\n```\nSTDERR:\n```\n{result.stderr}\n```"
    except subprocess.TimeoutExpired: 
        return "Execution timed out (>10s)."

def web_search_func(query: str, cache_func) -> str:
    """Performs a web search using Tavily and returns a compilation of results."""
    if not query or not query.strip():
        return "Error: Empty search query"
    
    key = f"web:{query}"
    try:
        results = cache_func(key, lambda: TavilySearchResults(max_results=5).invoke(query))
        if not results:
            return "No search results found"
        
        formatted_results = []
        for res in results:
            if isinstance(res, dict) and 'url' in res and 'content' in res:
                formatted_results.append(f"Source: {res['url']}\nContent: {res['content']}")
        
        return "\n\n---\n\n".join(formatted_results) if formatted_results else "No valid results found"
    except Exception as e:
        return f"Search error: {e}"

def wiki_search_func(query: str, cache_func) -> str:
    """Searches Wikipedia and returns the top 2 results."""
    if not query or not query.strip():
        return "Error: Empty search query"
    
    key = f"wiki:{query}"
    try:
        docs = cache_func(key, lambda: WikipediaLoader(
            query=query, 
            load_max_docs=2, 
            doc_content_chars_max=2000
        ).load())
        
        if not docs:
            return "No Wikipedia articles found"
        
        return "\n\n---\n\n".join([
            f"Source: {d.metadata.get('source', 'Unknown')}\n\n{d.page_content}" 
            for d in docs
        ])
    except Exception as e:
        return f"Wikipedia search error: {e}"

def arxiv_search_func(query: str, cache_func) -> str:
    """Searches Arxiv for scientific papers and returns the top 2 results."""
    if not query or not query.strip():
        return "Error: Empty search query"
    
    key = f"arxiv:{query}"
    try:
        docs = cache_func(key, lambda: ArxivLoader(query=query, load_max_docs=2).load())
        
        if not docs:
            return "No Arxiv papers found"
        
        return "\n\n---\n\n".join([
            f"Source: {d.metadata.get('source', 'Unknown')}\n"
            f"Published: {d.metadata.get('Published', 'Unknown')}\n"
            f"Title: {d.metadata.get('Title', 'Unknown')}\n\n"
            f"Summary:\n{d.page_content}" 
            for d in docs
        ])
    except Exception as e:
        return f"Arxiv search error: {e}"

@tool
def analyze_task_and_reason(task_description: str) -> str:
    """
    Analyzes the task and provides reasoning about what approach to take.
    This tool helps determine what other tools might be needed.
    """
    analysis = {
        "task_type": "unknown",
        "has_image": False,
        "needs_search": False,
        "needs_computation": False,
        "approach": "Direct answer"
    }
    
    task_lower = task_description.lower()
    
    # Check for image-related content
    if any(keyword in task_lower for keyword in [
        'image', 'picture', 'photo', 'visual', 'see in', 'shown in', 
        'attachment analysis', 'url:', 'http', '.jpg', '.png', '.gif'
    ]):
        analysis["has_image"] = True
        analysis["task_type"] = "image_analysis"
        analysis["approach"] = "Process image with vision model, then analyze content"
    
    # Check for search needs
    if any(keyword in task_lower for keyword in [
        'current', 'recent', 'latest', 'news', 'today', 'what is', 
        'who is', 'when did', 'research', 'find information'
    ]):
        analysis["needs_search"] = True
        if analysis["task_type"] == "unknown":
            analysis["task_type"] = "information_search"
            analysis["approach"] = "Search for current information"
    
    # Check for computation needs
    if any(keyword in task_lower for keyword in [
        'calculate', 'compute', 'math', 'formula', 'equation', 
        'algorithm', 'code', 'program', 'python'
    ]):
        analysis["needs_computation"] = True
        if analysis["task_type"] == "unknown":
            analysis["task_type"] = "computation"
            analysis["approach"] = "Use Python for calculations"
    
    reasoning = f"""TASK ANALYSIS COMPLETE:

Task Type: {analysis['task_type']}
Has Image: {analysis['has_image']}
Needs Search: {analysis['needs_search']}  
Needs Computation: {analysis['needs_computation']}

RECOMMENDED APPROACH: {analysis['approach']}

REASONING:
- If this involves an image, I should process it directly with my vision capabilities
- If this needs current information, I should use web search or Wikipedia
- If this needs calculations, I should use the Python tool
- I should always provide a comprehensive final answer

NEXT STEPS: Proceed with the identified approach and use appropriate tools."""
    
    return reasoning

# ----------------------------------------------------------
# Section 3: SYSTEM PROMPT
# ----------------------------------------------------------
SYSTEM_PROMPT = """You are an expert multimodal AI assistant with vision capabilities and access to various tools.

**CORE CAPABILITIES:**
1. **Vision Processing**: You can directly process and analyze images from URLs
2. **Web Search**: Access current information via web search and Wikipedia
3. **Computation**: Execute Python code for calculations and data processing
4. **Research**: Search academic papers and retrieve similar examples

**CRITICAL WORKFLOW:**
1. **ANALYZE FIRST**: Always start by using the 'analyze_task_and_reason' tool to understand what you're being asked to do
2. **PROCESS IMAGES DIRECTLY**: When you encounter image URLs, process them directly with your vision model - DO NOT use separate image tools
3. **USE TOOLS STRATEGICALLY**: Based on your analysis, use appropriate tools (web search, Python, etc.)
4. **VALIDATE PARAMETERS**: Always check that you're passing correct parameters to tools
5. **SYNTHESIZE**: Combine all information into a comprehensive answer

**IMAGE HANDLING:**
- You have native vision capabilities - process image URLs directly
- Look for image URLs in the task description
- When you see an image URL, examine it carefully and describe what you see
- Relate your visual observations to the question being asked

**TOOL USAGE RULES:**
- Always use 'analyze_task_and_reason' first to plan your approach
- Use web_search for current events, factual information, or research
- Use python_repl for calculations, data processing, or code execution
- Use wiki_search for encyclopedic information
- Use arxiv_search for academic/scientific papers
- Use retrieve_examples for similar solved problems

**OUTPUT FORMAT:**
Always end your response with your answer clearly stated on the last line.

**PARAMETER VALIDATION:**
- Check that search queries are meaningful and specific
- Ensure Python code is safe and well-formed
- Verify image URLs are accessible before processing
"""

# ----------------------------------------------------------
# Section 4: Factory Function for Agent Executor
# ----------------------------------------------------------
def create_agent_executor(provider: str = "groq"):
    """
    Factory function to create and compile the LangGraph agent executor.
    """
    print(f"Initializing agent with provider: {provider}")

    # Step 1: Initialize LLM with vision capabilities
    if provider == "groq":
        # Use Groq's vision-capable model
        try:
            llm = ChatGroq(
                model_name="meta-llama/llama-4-maverick-17b-128e-instruct" # Vision-capable model
            )
            print("Initialized Groq LLM with vision capabilities")
        except Exception as e:
            print(f"Error initializing Groq: {e}")
            raise
    else:
        raise ValueError(f"Provider '{provider}' not supported in this version")

    # Step 2: Build Retriever (if metadata exists)
    embeddings = HuggingFaceEmbeddings(model_name=EMBED_MODEL)
    if FAISS_CACHE.exists():
        with open(FAISS_CACHE, "rb") as f: 
            vector_store = pickle.load(f)
            print("Loaded existing FAISS index")
    else:
        if JSONL_PATH.exists():
            docs = []
            with open(JSONL_PATH, "rt", encoding="utf-8") as f:
                for line in f:
                    rec = json.loads(line)
                    docs.append(Document(
                        page_content=f"Question: {rec['Question']}\n\nFinal answer: {rec['Final answer']}", 
                        metadata={"source": rec["task_id"]}
                    ))
            vector_store = FAISS.from_documents(docs, embeddings)
            with open(FAISS_CACHE, "wb") as f: 
                pickle.dump(vector_store, f)
            print(f"Created new FAISS index with {len(docs)} documents")
        else:
            # Create minimal vector store
            docs = [Document(page_content="Sample document", metadata={"source": "sample"})]
            vector_store = FAISS.from_documents(docs, embeddings)
            print("Created minimal FAISS index")
    
    retriever = vector_store.as_retriever(search_kwargs={"k": RETRIEVER_K})
    
    # Step 3: Create tools list
    tools_list = [
        analyze_task_and_reason,
        Tool(
            name="web_search", 
            func=functools.partial(web_search_func, cache_func=cached_get), 
            description="Search the web for current information. Use specific, focused queries."
        ),
        Tool(
            name="wiki_search", 
            func=functools.partial(wiki_search_func, cache_func=cached_get), 
            description="Search Wikipedia for encyclopedic information."
        ),
        Tool(
            name="arxiv_search", 
            func=functools.partial(arxiv_search_func, cache_func=cached_get), 
            description="Search Arxiv for academic papers and research."
        ),
        python_repl,
        create_retriever_tool(
            retriever=retriever, 
            name="retrieve_examples", 
            description="Retrieve similar solved examples from the knowledge base."
        ),
    ]
    
    llm_with_tools = llm.bind_tools(tools_list)

    # Step 4: Define Graph Nodes
    def assistant_node(state: MessagesState):
        """Main assistant node that processes user input and tool responses."""
        messages = [SystemMessage(content=SYSTEM_PROMPT)] + state["messages"]
        try:
            result = llm_with_tools.invoke(messages)
            return {"messages": [result]}
        except Exception as e:
            error_msg = f"LLM Error: {e}"
            print(error_msg)
            return {"messages": [AIMessage(content=f"I encountered an error: {error_msg}")]}

    def tools_node_wrapper(state: MessagesState):
        """Wrapper for tool execution with error handling."""
        try:
            tool_node = ToolNode(tools_list)
            return tool_node.invoke(state)
        except Exception as e:
            error_msg = f"Tool execution error: {e}"
            print(error_msg)
            return {"messages": [AIMessage(content=error_msg)]}

    # Step 5: Build Graph
    builder = StateGraph(MessagesState)
    builder.add_node("assistant", assistant_node)
    builder.add_node("tools", tools_node_wrapper)
    
    builder.add_edge(START, "assistant")
    builder.add_conditional_edges(
        "assistant", 
        tools_condition, 
        {"tools": "tools", "__end__": "__end__"}
    )
    builder.add_edge("tools", "assistant")

    agent_executor = builder.compile()
    print("Agent Executor created successfully with vision capabilities")
    return agent_executor