Discover Connected Crimes with Web Search (Python/LangGraph)

Overview

In Exercise 05, you learned to search internal documents using the Grounding Service. Your Evidence Analyst node can now retrieve factual evidence from security logs, bank records, and termination letters without hallucination.

But investigations don't happen in a vacuum. What if similar crimes happened elsewhere? What if your suspects have public criminal records? What if this heist is part of a larger criminal network?

In this exercise, you'll add an Intelligence Researcher node that uses Perplexity's sonar-pro model to gather external intelligence from the web.

Understand Web Search with Sonar-Pro

Why Do We Need Web Search?

Your current investigation system is powerful but limited to internal data:

What You Can Do Now	What You Can't Do Yet
✅ Search museum's internal evidence documents	❌ Search public criminal databases
✅ Retrieve bank records, security logs	❌ Find similar crimes in other cities
✅ Analyze suspects based on internal evidence	❌ Check if suspects have public criminal records
✅ Ground responses in factual documents	❌ Discover criminal network connections
✅ Avoid hallucination for internal data	❌ Monitor online art markets for stolen items

Document Grounding vs. Web Search

Aspect	Document Grounding (Exercise 05)	Web Search (This Exercise)
Data Source	Internal documents (pre-uploaded)	Real-time web information
Coverage	Organization-specific evidence	Global public information
Freshness	Historical documents (static)	Current information (updated constantly)
Use Cases	Internal logs, private records, policies	News, criminal records, pattern analysis
Search Method	Vector similarity (semantic)	Web search + LLM synthesis
Tool Function	`call_grounding_service()`	`call_sonar_pro_search()`
When to Use	"What does our evidence say?"	"What do public records show?"

🎯 Key Insight: These are not alternatives — they work together!

How Sonar-Pro Works in LangGraph

In the Python-LangGraph track, tools are plain functions that nodes call directly — there is no @tool decorator. This is deterministic orchestration: the graph decides when to call the tool, not the LLM. This is different from CrewAI where the agent autonomously decides to use tools.

# Plain function — no @tool decorator needed
def call_sonar_pro_search(query: str) -> str:
    from litellm import completion
    response = completion(
        model="sap/sonar-pro",  # Perplexity via SAP AI Core
        messages=[...]
    )
    return response.choices[0].message.content

# Node calls it directly — the graph controls the flow
def intelligence_researcher_node(state: AgentState) -> dict:
    result = call_sonar_pro_search("Marcus Chen criminal record")
    return {"intelligence_report": result}

⚠️ Before continuing: Verify that sonar-pro is available in your SAP AI Core resource group. Go to SAP AI Launchpad → Generative AI Hub → Models and confirm you can see a Perplexity sonar-pro deployment. If it is not listed, ask your workshop instructor to enable it.

Add The Web Search Tool

Step 1: Add the `call_sonar_pro_search` Function

👉 Open /project/Python-LangGraph/starter-project/investigator_graph.py

👉 Add this function after call_grounding_service and before the model = ChatLiteLLM(...) line:

def call_sonar_pro_search(query: str) -> str:
    """Search the web using Perplexity's sonar-pro model for real-time intelligence."""
    from litellm import completion
    try:
        response = completion(
            model="sap/sonar-pro",
            messages=[
                {
                    "role": "system",
                    "content": "You are a web search assistant specializing in criminal intelligence. Search for accurate, recent information and always provide source citations with URLs and dates."
                },
                {
                    "role": "user",
                    "content": query
                }
            ],
            temperature=0.2,
        )
        return response.choices[0].message.content
    except Exception as e:
        return f"Error calling sonar-pro web search: {str(e)}"

💡 Why no @tool decorator?

In the LangGraph track, tool functions are called directly by nodes — the graph structure controls when tools are called, not the LLM. This is called deterministic tool orchestration. Compare this to CrewAI, where agents autonomously decide to call tools.

Extend the Agent State

Step 2: Add `intelligence_report` to `AgentState`

The LangGraph AgentState is a TypedDict that defines all the data flowing through the graph. You need to add a field for the intelligence report.

👉 Find the AgentState class in investigator_graph.py and add intelligence_report:

class AgentState(TypedDict):
    payload: dict
    suspect_names: str
    appraisal_result: Optional[str]
    evidence_analysis: Optional[str]
    intelligence_report: Optional[str]   # ← NEW
    final_conclusion: Optional[str]
    messages: list

💡 How LangGraph state works: Each node receives the full AgentState and returns a partial dict with only the fields it updated. LangGraph merges these partial updates into the shared state. This is why intelligence_researcher_node can return just {"intelligence_report": ...} — LangGraph carries the other fields forward automatically.

Add The Intelligence Researcher Node

Step 3: Add the Node Function

👉 Add this function after evidence_analyst_node and before lead_detective_node:

def intelligence_researcher_node(state: AgentState) -> dict:
    print("\n🔍 Intelligence Researcher starting web search...")

    try:
        suspects = [s.strip() for s in state["suspect_names"].split(",")]
        intelligence_results = []

        for suspect in suspects:
            print(f"  Searching public records for: {suspect}")
            query = f"{suspect} criminal record art theft security technician Europe background check"
            result = call_sonar_pro_search(query)
            intelligence_results.append(f"Background check for {suspect}:\n{result}")

        print("  Searching for similar art theft incidents...")
        pattern_query = "museum art theft insider job no forced entry Europe similar incidents criminal network"
        pattern_result = call_sonar_pro_search(pattern_query)
        intelligence_results.append(f"Similar Art Theft Patterns:\n{pattern_result}")

        intelligence_report = (
            "Intelligence Research Complete:\n\n" + "\n\n".join(intelligence_results) +
            f"\n\nSummary: Conducted OSINT research on all suspects and identified similar crime patterns"
        )

        print("✅ Intelligence research complete")

        return {
            "intelligence_report": intelligence_report,
            "messages": state["messages"] + [{"role": "assistant", "content": intelligence_report}],
        }
    except Exception as e:
        error_msg = f"Error during intelligence research: {e}"
        print(f"❌ {error_msg}")
        return {
            "intelligence_report": error_msg,
            "messages": state["messages"] + [{"role": "assistant", "content": error_msg}],
        }

Add the Agent Prompt

Step 4: Add `WEB_RESEARCHER_AGENT` to `config/agents.py`

👉 Open /project/Python-LangGraph/starter-project/config/agents.py

👉 Add this constant after EVIDENCE_ANALYST_AGENT:

WEB_RESEARCHER_AGENT = {
    "name": "intelligence_researcher_agent",
    "prompt": (
        "You are an Open-Source Intelligence (OSINT) specialist. "
        "Search the web for criminal records, similar art thefts, and suspect backgrounds. "
        "Focus on public information: news articles, court records, criminal databases. "
        "Always provide source citations with URLs and dates."
    ),
}

Update the Lead Detective Prompt

Step 5: Extend `_lead_detective_prompt`

The Lead Detective now synthesizes three sources instead of two. Update _lead_detective_prompt in config/agents.py:

def _lead_detective_prompt(appraisal_result: str, evidence_analysis: str, intelligence_report: str, suspect_names: str) -> str:
    return (
        "You are the Lead Detective coordinating an art theft investigation. "
        "You have received the following information from your team:\n\n"
        f"1. INSURANCE APPRAISAL:\n{appraisal_result}\n\n"
        f"2. EVIDENCE ANALYSIS (Internal Documents):\n{evidence_analysis}\n\n"
        f"3. INTELLIGENCE REPORT (Web Search):\n{intelligence_report}\n\n"
        f"4. SUSPECTS: {suspect_names}\n\n"
        "Based on all the evidence and analysis, you MUST:\n"
        "- Name the most likely thief and explain the evidence supporting that conclusion\n"
        "- Consider both internal evidence and external criminal patterns\n"
        "- Note any alibis or evidence that clears the other suspects\n"
        "- State the total insured value of the stolen goods\n"
        "- Assess whether this is an isolated incident or part of a larger criminal network\n"
        "- Provide a comprehensive summary of the case."
    )

Rewire the Graph

Step 6: Add the Node and Update Edges

👉 Find the build_graph() function in investigator_graph.py and update it:

def build_graph():
    workflow = StateGraph(AgentState)

    workflow.add_node("appraiser", appraiser_node)
    workflow.add_node("evidence_analyst", evidence_analyst_node)
    workflow.add_node("intelligence_researcher", intelligence_researcher_node)  # ← NEW
    workflow.add_node("lead_detective", lead_detective_node)
    workflow.add_edge(START, "appraiser")
    workflow.add_edge("appraiser", "evidence_analyst")
    workflow.add_edge("evidence_analyst", "intelligence_researcher")   # ← CHANGED
    workflow.add_edge("intelligence_researcher", "lead_detective")     # ← NEW
    workflow.add_edge("lead_detective", END)

    return workflow.compile()

The new graph structure:

flowchart TD
    START --> A[appraiser_node]
    A --> B[evidence_analyst_node]
    B --> C[intelligence_researcher_node]
    C --> D[lead_detective_node]
    D --> END

Update Lead Detective Node Call

Step 7: Pass `intelligence_report` to the Prompt

In investigator_graph.py, find lead_detective_node and update the LEAD_DETECTIVE["prompt"] call to include intelligence_report:

response = model.invoke([
    SystemMessage(content=LEAD_DETECTIVE["prompt"](
        state["appraisal_result"] or "No appraisal result available",
        state["evidence_analysis"] or "No evidence analysis available",
        state.get("intelligence_report") or "No intelligence report available",
        state["suspect_names"],
    )),
    HumanMessage(content="Analyze all the evidence and identify the culprit. Provide a detailed conclusion."),
])

Update `main.py` and `server.py`

Step 8: Initialize the New State Field

👉 Open /project/Python-LangGraph/starter-project/main.py

👉 Add "intelligence_report": None to the initial state dict:

result = investigator_graph.invoke({
    "payload": payload,
    "suspect_names": "Sophie Dubois, Marcus Chen, Viktor Petrov",
    "appraisal_result": None,
    "evidence_analysis": None,
    "intelligence_report": None,   # ← NEW
    "final_conclusion": None,
    "messages": [],
})

👉 Also add a print section for the intelligence report after the Evidence Analysis section:

print("\n" + "="*50)
print("Intelligence Report:")
print("="*50)
print(result["intelligence_report"] or "(not set)")

💡 server.py also constructs the initial state for investigator_graph.ainvoke(...). Add "intelligence_report": None there too — find the dict and add the field.

Run Your Enhanced Investigation

👉 Run the investigation from the starter-project folder:

macOS / Linux / BAS

python3 ./project/Python-LangGraph/starter-project/main.py

Windows (PowerShell)

python .\project\Python-LangGraph\starter-project\main.py

⏱️ Expected output — you should see all four nodes fire in order:

🔍 Appraiser Agent starting...
✅ Appraisal complete

🔍 Evidence Analyst starting...
✅ Evidence analysis complete

🔍 Intelligence Researcher starting web search...
  Searching public records for: Sophie Dubois
  Searching public records for: Marcus Chen
  Searching public records for: Viktor Petrov
  Searching for similar art theft incidents...
✅ Intelligence research complete

🔍 Lead Detective analyzing all findings...
✅ Investigation complete

Key Takeaways

LangGraph tool functions are plain functions — no @tool decorator needed. Nodes call them directly.
State carries data between nodes — intelligence_report is set by the Intelligence Researcher node and automatically available to the Lead Detective node via AgentState.
Deterministic orchestration — the graph structure controls the flow, not the LLM. Every run follows the same appraiser → evidence_analyst → intelligence_researcher → lead_detective path.
Sonar-Pro is called through the same litellm.completion() API as other models, just with model="sap/sonar-pro".

Next Steps

You now have a complete intelligence gathering system with:

✅ Structured data predictions (RPT-1)
✅ Internal document search (Grounding Service)
✅ External web intelligence (Sonar-Pro)

In the next exercise, you'll finalize the Lead Detective prompt to synthesize all three sources and solve the museum art theft mystery.

Troubleshooting

Issue: ModuleNotFoundError: No module named 'litellm'

Solution: LiteLLM should already be installed from Exercise 02. If not:

macOS / Linux / BAS
```
pip install litellm==1.82.6
```

Issue: Error: Model sap/sonar-pro not found or 404

Solution: Verify that sonar-pro is deployed in your SAP AI Core resource group. Check SAP AI Launchpad → Generative AI Hub → Models.

Issue: KeyError: 'intelligence_report'

Solution: Ensure you added intelligence_report: Optional[str] to the AgentState TypedDict AND initialized it to None in main.py's initial state dict.

Issue: TypeError: _lead_detective_prompt() takes 3 positional arguments but 4 were given

Solution: You updated the lead_detective_node call but forgot to update _lead_detective_prompt's signature in config/agents.py (or vice versa). Make sure both accept intelligence_report as the third parameter.

Issue: Intelligence Researcher fires but Lead Detective ignores the report

Solution: Check that _lead_detective_prompt actually interpolates {intelligence_report} in its return string, and that lead_detective_node passes state.get("intelligence_report") as the third argument.

Resources

Next exercise

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Discover Connected Crimes with Web Search (Python/LangGraph)

Overview

Understand Web Search with Sonar-Pro

Why Do We Need Web Search?

Document Grounding vs. Web Search

How Sonar-Pro Works in LangGraph

Add The Web Search Tool

Step 1: Add the `call_sonar_pro_search` Function

Extend the Agent State

Step 2: Add `intelligence_report` to `AgentState`

Add The Intelligence Researcher Node

Step 3: Add the Node Function

Add the Agent Prompt

Step 4: Add `WEB_RESEARCHER_AGENT` to `config/agents.py`

Update the Lead Detective Prompt

Step 5: Extend `_lead_detective_prompt`

Rewire the Graph

Step 6: Add the Node and Update Edges

Update Lead Detective Node Call

Step 7: Pass `intelligence_report` to the Prompt

Update `main.py` and `server.py`

Step 8: Initialize the New State Field

Run Your Enhanced Investigation

Key Takeaways

Next Steps

Troubleshooting

Resources

Uh oh!

FilesExpand file tree

06-discover-connected-crimes.md

Latest commit

History

06-discover-connected-crimes.md

File metadata and controls

Discover Connected Crimes with Web Search (Python/LangGraph)

Overview

Understand Web Search with Sonar-Pro

Why Do We Need Web Search?

Document Grounding vs. Web Search

How Sonar-Pro Works in LangGraph

Add The Web Search Tool

Step 1: Add the call_sonar_pro_search Function

Extend the Agent State

Step 2: Add intelligence_report to AgentState

Add The Intelligence Researcher Node

Step 3: Add the Node Function

Add the Agent Prompt

Step 4: Add WEB_RESEARCHER_AGENT to config/agents.py

Update the Lead Detective Prompt

Step 5: Extend _lead_detective_prompt

Rewire the Graph

Step 6: Add the Node and Update Edges

Update Lead Detective Node Call

Step 7: Pass intelligence_report to the Prompt

Update main.py and server.py

Step 8: Initialize the New State Field

Run Your Enhanced Investigation

Key Takeaways

Next Steps

Troubleshooting

Resources

Step 1: Add the `call_sonar_pro_search` Function

Step 2: Add `intelligence_report` to `AgentState`

Step 4: Add `WEB_RESEARCHER_AGENT` to `config/agents.py`

Step 5: Extend `_lead_detective_prompt`

Step 7: Pass `intelligence_report` to the Prompt

Update `main.py` and `server.py`