diff --git a/.github/copilot-instructions.md b/.github/copilot-instructions.md
new file mode 100644
index 0000000..2aaf0a3
--- /dev/null
+++ b/.github/copilot-instructions.md
@@ -0,0 +1,62 @@
+# SAP CodeJam: Code-Based AI Agents
+
+Hands-on workshop for building multi-agent AI systems on SAP BTP using **CrewAI** and **LiteLLM** connected to **SAP Generative AI Hub** (AI Core). The scenario: investigate an art heist by orchestrating specialized agents that analyse evidence, appraise stolen items, and identify the culprit.
+
+## Architecture
+
+Three-agent sequential crew in `project/Python/`:
+
+| Agent | Tool | Purpose |
+|---|---|---|
+| Appraiser | `call_rpt1()` | Predict item categories & insurance values via SAP-RPT-1 ML model |
+| Evidence Analyst | `call_grounding_service()` | RAG queries over evidence documents via SAP Grounding Service |
+| Lead Detective | _(none)_ | Synthesise findings and name the culprit |
+
+Key files in `project/Python/solution/`:
+- `investigator_crew.py` — `@CrewBase` class; agents and tasks defined with `@agent`, `@task`, `@crew` decorators; tools use `@tool("name")`
+- `main.py` — entry point; calls `crew().kickoff(inputs={...})`
+- `rpt_client.py` — OAuth2 client for SAP-RPT-1 (client-credentials grant, Bearer auth)
+- `payload.py` — structured art-item data with `[PREDICT]` placeholders for RPT-1
+- `config/agents.yaml`, `config/tasks.yaml` — YAML definitions (method names must match decorator names)
+
+Evidence documents (plain text, loaded into the Grounding Service pipeline) are in `exercises/data/documents/`.
+
+## Build & Run
+
+```bash
+# Activate virtual env (Windows)
+.\env\Scripts\Activate.ps1
+
+# Install dependencies (only needed once)
+pip install litellm crewai python-dotenv
+
+# Run the crew
+python main.py
+```
+
+Requires a `.env` file in `project/Python/starter-project/` (copy structure from the exercise docs):
+
+```
+AICORE_CLIENT_ID=
+AICORE_CLIENT_SECRET=
+AICORE_AUTH_URL=
+RPT1_DEPLOYMENT_URL=
+AICORE_RESOURCE_GROUP=
+AICORE_BASE_URL=
+```
+
+## Conventions
+
+- **CrewAI YAML config**: agent/task names in `agents.yaml` / `tasks.yaml` must exactly match the Python method names decorated with `@agent` / `@task`.
+- **Tool pattern**: tools are plain functions decorated with `@tool("Descriptive Name")`. Return error messages as strings so the LLM can handle failures gracefully.
+- **LLM model strings**: use `sap/<model-name>` format (e.g. `sap/gpt-4o`) matching deployments in SAP AI Launchpad.
+- **Process**: always `Process.sequential` — tasks pass outputs as context to the next task in order.
+- **RPT-1 payload**: `[PREDICT]` string is the placeholder for values to be inferred; schema (dtype, categories, value ranges) must be exact.
+
+## Pitfalls
+
+- **Hardcoded Grounding pipeline ID** in `call_grounding_service()` — replace with your own vector DB pipeline ID from SAP AI Launchpad before running.
+- **Token refresh**: `RPT1Client` fetches the OAuth token once at init; long-running crews may hit expiry — re-instantiate if needed.
+- **No `.env` validation** at startup — credential errors only surface on the first API call.
+- **YAML / decorator name mismatch** causes silent CrewAI failures with no clear error message.
+- **Grounding pipeline must be pre-loaded** with the evidence documents; empty pipelines return no results and agents will hallucinate.
diff --git a/.github/instructions/python-sap-tools.instructions.md b/.github/instructions/python-sap-tools.instructions.md
new file mode 100644
index 0000000..987aef8
--- /dev/null
+++ b/.github/instructions/python-sap-tools.instructions.md
@@ -0,0 +1,125 @@
+---
+description: "Use when writing Python code for SAP AI Core integrations: CrewAI agents, LiteLLM models, SAP-RPT-1 API calls, Grounding Service queries, OAuth2 token handling. Covers tool patterns, model string format, error handling, and YAML/decorator conventions."
+applyTo: "**/*.py"
+---
+
+# Python SAP Tools Conventions
+
+## LLM Model Strings
+
+Use the `sap/<model-name>` format matching the deployment name in SAP AI Launchpad:
+
+```python
+llm = LLM(model="sap/gpt-4o")
+```
+
+Never use bare provider strings like `"gpt-4o"` or `"openai/gpt-4o"` for SAP AI Core deployments.
+
+## CrewAI Agent & Task Pattern
+
+Agents and their tasks are defined in YAML; the Python method names **must exactly match** the YAML keys:
+
+```python
+@CrewBase
+class MyCrew():
+    agents_config = "config/agents.yaml"
+    tasks_config  = "config/tasks.yaml"
+
+    @agent
+    def appraiser_agent(self) -> Agent:   # key in agents.yaml: appraiser_agent
+        return Agent(config=self.agents_config["appraiser_agent"], tools=[call_rpt1])
+
+    @task
+    def appraisal_task(self) -> Task:     # key in tasks.yaml: appraisal_task
+        return Task(config=self.tasks_config["appraisal_task"])
+
+    @crew
+    def crew(self) -> Crew:
+        return Crew(agents=self.agents, tasks=self.tasks, process=Process.sequential)
+```
+
+Always use `Process.sequential`; task outputs are automatically passed as context to subsequent tasks.
+
+## Tool Pattern
+
+Tools are module-level functions decorated with `@tool`. Return error messages as plain strings so the LLM can recover gracefully — never raise exceptions out of a tool:
+
+```python
+from crewai.tools import tool
+
+@tool("call_rpt1")
+def call_rpt1(payload: dict) -> str:
+    """Docstring is the tool description shown to the agent."""
+    try:
+        response = rpt1_client.post_request(json_payload=payload)
+        if response.status_code == 200:
+            return json.dumps(response.json(), indent=2)
+        return f"Error {response.status_code}: {response.text}"
+    except Exception as e:
+        return f"Error calling RPT-1: {str(e)}"
+```
+
+## OAuth2 Client-Credentials Pattern (SAP AI Core)
+
+```python
+import os, requests
+
+data = {
+    "grant_type": "client_credentials",
+    "client_id": os.getenv("AICORE_CLIENT_ID"),
+    "client_secret": os.getenv("AICORE_CLIENT_SECRET"),
+}
+headers = {"Content-Type": "application/x-www-form-urlencoded"}
+resp = requests.post(os.getenv("AICORE_AUTH_URL"), data=data, headers=headers, timeout=30)
+resp.raise_for_status()
+token = resp.json()["access_token"]
+```
+
+Subsequent requests use `"Authorization": f"Bearer {token}"` and `"AI-Resource-Group": resource_group`.
+
+> **Token expiry**: the token is fetched once at init. For long-running crews, re-instantiate the client to refresh.
+
+## Grounding Service Query Pattern
+
+```python
+from gen_ai_hub.document_grounding.client import RetrievalAPIClient
+from gen_ai_hub.document_grounding.models.retrieval import RetrievalSearchInput, RetrievalSearchFilter
+from gen_ai_hub.orchestration.models.document_grounding import DataRepositoryType
+
+@tool("call_grounding_service")
+def call_grounding_service(user_question: str) -> str:
+    """Query evidence documents via SAP Grounding Service."""
+    client = RetrievalAPIClient()
+    search_filter = RetrievalSearchFilter(
+        id="vector",
+        dataRepositoryType=DataRepositoryType.VECTOR.value,
+        dataRepositories=["<YOUR_PIPELINE_ID>"],  # Replace with pipeline ID from SAP AI Launchpad
+        searchConfiguration={"maxChunkCount": 5},
+    )
+    search_input = RetrievalSearchInput(query=user_question, filters=[search_filter])
+    response = client.search(search_input)
+    return json.dumps(response.model_dump(), indent=2)
+```
+
+Replace `<YOUR_PIPELINE_ID>` with the vector DB pipeline ID from SAP AI Launchpad before running.
+
+## RPT-1 Payload
+
+Use the string `"[PREDICT]"` as the placeholder for values the model should infer. The schema (dtype, categories, value ranges) must match the deployed model exactly. Pass the payload dict directly from `main.py` inputs via the crew kickoff:
+
+```python
+crew.kickoff(inputs={"payload": payload, "query": "Who stole the art?"})
+```
+
+## Environment Variables
+
+Load `.env` once at the top of the entry-point module, before instantiating any client:
+
+```python
+from pathlib import Path
+from dotenv import load_dotenv
+
+load_dotenv(dotenv_path=Path(__file__).parent / ".env")
+```
+
+Required keys: `AICORE_CLIENT_ID`, `AICORE_CLIENT_SECRET`, `AICORE_AUTH_URL`, `RPT1_DEPLOYMENT_URL`, `AICORE_RESOURCE_GROUP`, `AICORE_BASE_URL`.
diff --git a/exercises/Python/01-setup-dev-space.md b/exercises/Python/01-setup-dev-space.md
index 7591836..92ba160 100644
--- a/exercises/Python/01-setup-dev-space.md
+++ b/exercises/Python/01-setup-dev-space.md
@@ -88,7 +88,14 @@ RPT1_DEPLOYMENT_URL="https://api.ai.prod.eu-central-1.aws.ml.hana.ondemand.com/v
 python3 --version
 ```
 
-> ⚠️ **Important**: CrewAI requires Python 3.10 or newer (up to 3.13). If your Python version is 3.9 or older, you'll need to install a compatible version. On macOS, you can use [Homebrew](https://brew.sh/) to install Python 3.11: `brew install python@3.11`, then use `python3.11` instead of `python3` in the commands below.
+> ⚠️ **Important**: CrewAI requires Python 3.10 or newer (up to 3.13). If your Python version is 3.9 or older, install a compatible version first.
+>
+> - **macOS**: Use [Homebrew](https://brew.sh/) to install Python 3.11: `brew install python@3.11`, then use `python3.11` in the commands below.
+> - **Linux**: Install Python 3.11 with your distro package manager, for example:
+>   - Ubuntu/Debian: `sudo apt update && sudo apt install python3.11 python3.11-venv`
+>   - Fedora/RHEL: `sudo dnf install python3.11`
+>   Then use `python3.11` in the commands below.
+> - **Windows**: Install Python 3.11 from the [official Python downloads page](https://www.python.org/downloads/windows/) and make sure **Add python.exe to PATH** is enabled during installation. Then use `python` (or `py -3.11`) in the commands below.
 
 👉 Create a virtual environment using the following command:
 
@@ -96,12 +103,43 @@ python3 --version
 python3 -m venv ~/projects/codejam-code-based-agents/env --upgrade-deps
 ```
 
+Or use the variant that matches your OS/shell if not in BAS:
+
+```bash
+# macOS / Linux
+python3 -m venv ~/projects/codejam-code-based-agents/env --upgrade-deps
+```
+
+```powershell
+# Windows (PowerShell)
+python -m venv .\env --upgrade-deps
+```
+
 👉 Activate the `env` virtual environment like this and make sure it is activated:
 
 ```bash
 source ~/projects/codejam-code-based-agents/env/bin/activate
 ```
 
+Use the activation command for your environment:
+
+```bash
+# macOS / Linux
+source ~/projects/codejam-code-based-agents/env/bin/activate
+```
+
+```powershell
+# Windows (PowerShell)
+.\env\Scripts\Activate.ps1
+```
+
+```cmd
+:: Windows (Command Prompt)
+.\env\Scripts\activate.bat
+```
+
+> ℹ️ If PowerShell blocks script execution, run `Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass` and then run `./env/Scripts/Activate.ps1` again.
+
 ![venv](/exercises/data/images/venv.png)
 
 👉 Install LiteLLM, CrewAI, and python-dotenv using the following `pip install` commands.
@@ -114,6 +152,6 @@ pip install litellm crewai python-dotenv
 
 ![bas-message](/exercises/data/images/virtual-env-python-bas-warning.png)
 
-## Let's start coding!
+## Let's start coding
 
 [Next exercise](02-build-a-basic-agent.md)
diff --git a/exercises/Python/02-build-a-basic-agent.md b/exercises/Python/02-build-a-basic-agent.md
index 0bdc7fc..59c3e04 100644
--- a/exercises/Python/02-build-a-basic-agent.md
+++ b/exercises/Python/02-build-a-basic-agent.md
@@ -109,10 +109,40 @@ if __name__ == "__main__":
 
 👉 Execute the crew with the basic agent:
 
-> ☝️ Make sure you're in the repository root directory (e.g., `codejam-code-based-agents-1`) when running this command. If you're already in the `starter-project` folder, use `python basic_agent.py` instead.
+> ☝️ Make sure you're in the repository root directory (e.g., `codejam-code-based-agents-1`) when running this command. If you're already in the `starter-project` folder, use the appropriate command for your OS.
+
+**From repository root:**
 
 ```bash
-python project/Python/starter-project/basic_agent.py
+# macOS / Linux / BAS
+python3 ./basic_agent.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\basic_agent.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\basic_agent.py
+```
+
+**From starter-project folder:**
+
+```bash
+# macOS / Linux
+python3 basic_agent.py
+```
+
+```powershell
+# Windows (PowerShell)
+python basic_agent.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python basic_agent.py
 ```
 
 You should see:
@@ -136,7 +166,7 @@ You created and ran a working AI agent that:
 
 The basic workflow is:
 
-```
+```text
 Task → Agent (Role/Goal/Backstory) → LLM Processing (GPT-4o) → Response → Output
 ```
 
@@ -167,7 +197,25 @@ In the following exercises, you will:
 
 **Issue**: `ModuleNotFoundError: No module named 'crewai'`
 
-- **Solution**: Ensure you're in the correct Python environment: `source venv/bin/activate` and run `pip install crewai litellm`
+- **Solution**: Ensure you're in the correct Python environment and run the install command.
+
+```bash
+# macOS / Linux - Activate environment and install
+source ~/projects/codejam-code-based-agents/env/bin/activate
+pip install crewai litellm
+```
+
+```powershell
+# Windows (PowerShell) - Activate environment and install
+.\env\Scripts\Activate.ps1
+pip install crewai litellm
+```
+
+```cmd
+# Windows (Command Prompt) - Activate environment and install
+.\env\Scripts\activate.bat
+pip install crewai litellm
+```
 
 ---
 
diff --git a/exercises/Python/03-add-your-first-tool.md b/exercises/Python/03-add-your-first-tool.md
index f1e9f19..616ab0f 100644
--- a/exercises/Python/03-add-your-first-tool.md
+++ b/exercises/Python/03-add-your-first-tool.md
@@ -289,10 +289,40 @@ if __name__ == "__main__":
 
 👉 Run your crew to test it.
 
-> ☝️ Make sure you're in the repository root directory (e.g., `codejam-code-based-agents`) when running this command. If you're already in the `starter-project` folder, use `python basic_agent.py` instead.
+> ☝️ Make sure you run from the starter-project folder, or use the full path from the repository root.
+
+**From repository root:**
+
+```bash
+# macOS / Linux
+python3 ./project/Python/starter-project/basic_agent.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\project\Python\starter-project\basic_agent.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\project\Python\starter-project\basic_agent.py
+```
+
+**From starter-project folder:**
 
 ```bash
-python project/Python/starter-project/basic_agent.py
+# macOS / Linux
+python3 basic_agent.py
+```
+
+```powershell
+# Windows (PowerShell)
+python basic_agent.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python basic_agent.py
 ```
 
 ☝️ You added an input variable to your agent but the agent is still not using a tool. Let's build the actual tool next.
@@ -541,10 +571,40 @@ RPT1_DEPLOYMENT_URL="https://api.ai.prod.eu-central-1.aws.ml.hana.ondemand.com/v
 
 👉 Run your crew to test it.
 
-> ☝️ Make sure you're in the repository root directory (e.g., `codejam-code-based-agents`) when running this command. If you're already in the `starter-project` folder, use `python basic_agent.py` instead.
+> ☝️ Make sure you run from the starter-project folder, or use the full path from the repository root.
+
+**From starter-project folder:**
 
 ```bash
-python project/Python/starter-project/basic_agent.py
+# macOS / Linux
+python3 basic_agent.py
+```
+
+```powershell
+# Windows (PowerShell)
+python basic_agent.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python basic_agent.py
+```
+
+**From repository root:**
+
+```bash
+# macOS / Linux
+python3 ./project/Python/starter-project/basic_agent.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\project\Python\starter-project\basic_agent.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\project\Python\starter-project\basic_agent.py
 ```
 
 👉 Understand the output of the agent using SAP-RPT-1 as a tool.
@@ -565,7 +625,7 @@ You extended your agent with:
 
 ### The Tool Flow
 
-```
+```text
 Agent Task → LLM Reasoning → Tool Decision → Tool Execution → Result → Agent Processing → LLM Reasoning → Output
 ```
 
diff --git a/exercises/Python/04-building-multi-agent-system.md b/exercises/Python/04-building-multi-agent-system.md
index 8c86be1..0a9a8c1 100644
--- a/exercises/Python/04-building-multi-agent-system.md
+++ b/exercises/Python/04-building-multi-agent-system.md
@@ -218,7 +218,7 @@ if __name__ == "__main__":
 
 ### Step 1: Adding a New Agent
 
-We also have a lot of evidence in our evidence database. You can check the documents that are part of the evidence [here](exercises/data/documents).
+We also have a lot of evidence in our evidence database. You can check the documents that are part of the evidence [here](/exercises/data/documents).
 
 To analyze the evidence and find all the information on our three suspects:
 
@@ -294,8 +294,38 @@ analyze_evidence_task:
 
 👉 Run your crew to test it.
 
+**From repository root:**
+
 ```bash
-python project/Python/starter-project/main.py
+# macOS / Linux
+python3 ./main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\main.py
+```
+
+**From starter-project folder:**
+
+```bash
+# macOS / Linux
+python3 main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python main.py
 ```
 
 > 💡 **Note:** The Evidence Analyst currently uses `call_rpt1` as a placeholder tool to make the code functional. This isn't the right tool for evidence analysis. You'll replace it with the `call_grounding_service` tool in Exercise 05 to give the agent proper access to evidence documents.
diff --git a/exercises/Python/05-add-the-grounding-service.md b/exercises/Python/05-add-the-grounding-service.md
index d478bc3..218fa0c 100644
--- a/exercises/Python/05-add-the-grounding-service.md
+++ b/exercises/Python/05-add-the-grounding-service.md
@@ -78,70 +78,36 @@ Before your agent can search documents, they must be prepared:
 
 When your agent asks a question, here's what happens:
 
+```mermaid
+flowchart TD
+    A["Agent Question: \"What evidence exists about Marcus Chen?\""]
+    B["1. Convert Query to Vector Embedding<br/>\"Marcus Chen evidence\" → [0.23, -0.45, 0.87, ...]"]
+    C["2. Search Vector Database (Similarity Search)<br/>Find document chunks with similar vectors<br/>(Cosine similarity scores 0.0 - 1.0)"]
+    D["3. Retrieve Top 5 Most Relevant Chunks<br/>✓ MARCUS_TERMINATION_LETTER.txt (score: 0.92)<br/>✓ SECURITY_LOG.txt - Marcus entries (score: 0.88)<br/>✓ BANK_RECORDS.txt - Marcus account (score: 0.85)<br/>✓ MARCUS_EXIT_LOG.txt (score: 0.83)<br/>✓ PHONE_RECORDS.txt - Marcus calls (score: 0.79)"]
+    E["Return to Agent"]
+
+    A --> B --> C --> D --> E
 ```
-┌─────────────────────────────────────────────────────────────┐
-│ Agent Question: "What evidence exists about Marcus Chen?"  │
-└─────────────────────────┬───────────────────────────────────┘
-                          ↓
-┌─────────────────────────────────────────────────────────────┐
-│ 1. Convert Query to Vector Embedding                        │
-│    "Marcus Chen evidence" → [0.23, -0.45, 0.87, ...]       │
-└─────────────────────────┬───────────────────────────────────┘
-                          ↓
-┌─────────────────────────────────────────────────────────────┐
-│ 2. Search Vector Database (Similarity Search)               │
-│    Find document chunks with similar vectors                │
-│    (Cosine similarity scores 0.0 - 1.0)                    │
-└─────────────────────────┬───────────────────────────────────┘
-                          ↓
-┌─────────────────────────────────────────────────────────────┐
-│ 3. Retrieve Top 5 Most Relevant Chunks                      │
-│    ✓ MARCUS_TERMINATION_LETTER.txt (score: 0.92)          │
-│    ✓ SECURITY_LOG.txt - Marcus entries (score: 0.88)      │
-│    ✓ BANK_RECORDS.txt - Marcus account (score: 0.85)      │
-│    ✓ MARCUS_EXIT_LOG.txt (score: 0.83)                    │
-│    ✓ PHONE_RECORDS.txt - Marcus calls (score: 0.79)       │
-└─────────────────────────┬───────────────────────────────────┘
-                          ↓
-                   Return to Agent
-```
+
+If Mermaid doesn't render in your viewer, see the plain-text version in [ASCII Fallbacks](#ascii-fallbacks).
 
 > ⚡ **Speed:** Vector search is incredibly fast—searches millions of documents!
 
 #### **Phase 3: Context-Enhanced Response**
 
-```
-┌─────────────────────────────────────────────────┐
-│ Retrieved Document Chunks (with text)           │
-│ ────────────────────────────────────────        │
-│ Chunk 1: "Marcus Chen was terminated on..."    │
-│ Chunk 2: "Security logs show Marcus accessed..."│
-│ Chunk 3: "Bank records indicate deposits of..." │
-└──────────────────┬──────────────────────────────┘
-                   ↓
-       Pass as Context to LLM
-                   ↓
-┌─────────────────────────────────────────────────┐
-│ LLM Prompt:                                     │
-│ "Based ONLY on these documents, answer:         │
-│  What evidence exists about Marcus Chen?        │
-│                                                  │
-│ Documents:                                      │
-│ [chunks inserted here]"                         │
-└──────────────────┬──────────────────────────────┘
-                   ↓
-┌─────────────────────────────────────────────────┐
-│ LLM generates answer grounded in facts:         │
-│                                                  │
-│ "According to MARCUS_TERMINATION_LETTER.txt,    │
-│  Marcus was fired on 2024-01-15 for             │
-│  'unauthorized access.' SECURITY_LOG.txt shows  │
-│  he entered secured areas 3 times after hours..." │
-└──────────────────┬──────────────────────────────┘
-                   ↓
-           Agent receives factual response
+```mermaid
+flowchart TD
+    A["Retrieved Document Chunks (with text)<br/>────────────────────────────────────────<br/>Chunk 1: \"Marcus Chen was terminated on...\"<br/>Chunk 2: \"Security logs show Marcus accessed...\"<br/>Chunk 3: \"Bank records indicate deposits of...\""]
+    B["Pass as Context to LLM"]
+    C["LLM Prompt:<br/>\"Based ONLY on these documents, answer:<br/>What evidence exists about Marcus Chen?<br/><br/>Documents:<br/>[chunks inserted here]\""]
+    D["LLM generates answer grounded in facts:<br/><br/>\"According to MARCUS_TERMINATION_LETTER.txt,<br/>Marcus was fired on 2024-01-15 for<br/>'unauthorized access.' SECURITY_LOG.txt shows<br/>he entered secured areas 3 times after hours...\""]
+    E["Agent receives factual response"]
+
+    A --> B --> C --> D --> E
 ```
 
+If Mermaid doesn't render in your viewer, see the plain-text version in [ASCII Fallbacks](#ascii-fallbacks).
+
 > 🎯 **Key Insight:** The LLM can **only** use information from the retrieved chunks. It can't make things up which is called hallucination!
 
 ### The Grounding Pipeline
@@ -421,8 +387,38 @@ def evidence_analyst_agent(self) -> Agent:
 
 👉 Run your crew to test the grounding service!
 
+**From repository root:**
+
+```bash
+# macOS / Linux
+python3 ./project/Python/starter-project/main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\project\Python\starter-project\main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\project\Python\starter-project\main.py
+```
+
+**From starter-project folder:**
+
 ```bash
-python project/Python/starter-project/main.py
+# macOS / Linux
+python3 main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python main.py
 ```
 
 Your Evidence Analyst should now search through actual evidence documents and cite specific sources (like "MARCUS_TERMINATION_LETTER.txt") instead of making up information!
@@ -443,7 +439,7 @@ You integrated a grounding service tool with your agent that:
 
 ### The Grounding Flow
 
-```
+```text
 User Query → LLM Reasoning → Agent Processing → Grounding Tool Call → Vector Search → Document Chunks → Agent Processing → LLM Reasoning → Output
 ```
 
@@ -514,6 +510,74 @@ In the following exercises, you will:
 
 ---
 
+## ASCII Fallbacks
+
+### Phase 2: Query Processing (ASCII)
+
+```text
+┌─────────────────────────────────────────────────────────────┐
+│ Agent Question: "What evidence exists about Marcus Chen?"  │
+└─────────────────────────┬───────────────────────────────────┘
+              ↓
+┌─────────────────────────────────────────────────────────────┐
+│ 1. Convert Query to Vector Embedding                        │
+│    "Marcus Chen evidence" → [0.23, -0.45, 0.87, ...]       │
+└─────────────────────────┬───────────────────────────────────┘
+              ↓
+┌─────────────────────────────────────────────────────────────┐
+│ 2. Search Vector Database (Similarity Search)               │
+│    Find document chunks with similar vectors                │
+│    (Cosine similarity scores 0.0 - 1.0)                    │
+└─────────────────────────┬───────────────────────────────────┘
+              ↓
+┌─────────────────────────────────────────────────────────────┐
+│ 3. Retrieve Top 5 Most Relevant Chunks                      │
+│    ✓ MARCUS_TERMINATION_LETTER.txt (score: 0.92)          │
+│    ✓ SECURITY_LOG.txt - Marcus entries (score: 0.88)      │
+│    ✓ BANK_RECORDS.txt - Marcus account (score: 0.85)      │
+│    ✓ MARCUS_EXIT_LOG.txt (score: 0.83)                    │
+│    ✓ PHONE_RECORDS.txt - Marcus calls (score: 0.79)       │
+└─────────────────────────┬───────────────────────────────────┘
+              ↓
+           Return to Agent
+```
+
+### Phase 3: Context-Enhanced Response (ASCII)
+
+```text
+┌─────────────────────────────────────────────────┐
+│ Retrieved Document Chunks (with text)           │
+│ ────────────────────────────────────────        │
+│ Chunk 1: "Marcus Chen was terminated on..."    │
+│ Chunk 2: "Security logs show Marcus accessed..."│
+│ Chunk 3: "Bank records indicate deposits of..." │
+└──────────────────┬──────────────────────────────┘
+           ↓
+       Pass as Context to LLM
+           ↓
+┌─────────────────────────────────────────────────┐
+│ LLM Prompt:                                     │
+│ "Based ONLY on these documents, answer:         │
+│  What evidence exists about Marcus Chen?        │
+│                                                  │
+│ Documents:                                      │
+│ [chunks inserted here]"                         │
+└──────────────────┬──────────────────────────────┘
+           ↓
+┌─────────────────────────────────────────────────┐
+│ LLM generates answer grounded in facts:         │
+│                                                  │
+│ "According to MARCUS_TERMINATION_LETTER.txt,    │
+│  Marcus was fired on 2024-01-15 for             │
+│  'unauthorized access.' SECURITY_LOG.txt shows  │
+│  he entered secured areas 3 times after hours..." │
+└──────────────────┬──────────────────────────────┘
+           ↓
+       Agent receives factual response
+```
+
+---
+
 ## Resources
 
 - [SAP AI Core Grounding Management](https://help.sap.com/docs/sap-ai-core/sap-ai-core-service-guide/document-grounding)
diff --git a/exercises/Python/06-solve-the-crime.md b/exercises/Python/06-solve-the-crime.md
index b8c4dc6..f88608c 100644
--- a/exercises/Python/06-solve-the-crime.md
+++ b/exercises/Python/06-solve-the-crime.md
@@ -119,8 +119,38 @@ result = InvestigatorCrew().crew().kickoff(inputs={
 
 👉 Run your crew to start the investigation!
 
+**From repository root:**
+
+```bash
+# macOS / Linux
+python3 ./project/Python/starter-project/main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\project\Python\starter-project\main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\project\Python\starter-project\main.py
+```
+
+**From starter-project folder:**
+
 ```bash
-python project/Python/starter-project/main.py
+# macOS / Linux
+python3 main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python main.py
 ```
 
 > ⏱️ **This may take 2-5 minutes** as your agents:
@@ -160,7 +190,41 @@ If your Lead Detective identifies the wrong suspect, you'll need to refine your
 - ❌ Avoid vague instructions like "solve the crime" without guidance
 - ❌ Don't assume agents know which evidence is most important
 
-👉 After updating prompts in the YAML files, run `python project/Python/starter-project/main.py` again
+👉 After updating prompts in the YAML files, run the crew again:
+
+**From repository root:**
+
+```bash
+# macOS / Linux
+python3 ./project/Python/starter-project/main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python .\project\Python\starter-project\main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python .\project\Python\starter-project\main.py
+```
+
+**From starter-project folder:**
+
+```bash
+# macOS / Linux
+python3 main.py
+```
+
+```powershell
+# Windows (PowerShell)
+python main.py
+```
+
+```cmd
+# Windows (Command Prompt)
+python main.py
+```
 
 👉 Verify the answer with the instructor
 
@@ -180,7 +244,7 @@ You created a complete multi-agent system where:
 
 ### The Investigation Flow
 
-```
+```text
 Lead Detective → Evidence Analysis → Grounding Search → Suspect Investigation
               ↓
          Loss Appraisal → RPT-1 Predictions → Value Determination
@@ -236,14 +300,15 @@ Congratulations on completing the CodeJam! You've successfully built a sophistic
 **Issue**: `AttributeError: 'NoneType' object has no attribute 'get'` when running main.py
 
 - **Solution**: This error occurs when YAML configuration has incorrect indentation. Check your `config/tasks.yaml` file and ensure all fields (`description`, `expected_output`, `agent`) are indented with **2 spaces** under each task name:
-  ```yaml
-  solve_crime:
-    description: > # ← Must be indented 2 spaces
-      Your task description
-    expected_output: > # ← Must be indented 2 spaces
-      Expected output
-    agent: lead_detective_agent # ← Must be indented 2 spaces
-  ```
+
+```yaml
+solve_crime:
+  description: > # ← Must be indented 2 spaces
+    Your task description
+  expected_output: > # ← Must be indented 2 spaces
+    Expected output
+  agent: lead_detective_agent # ← Must be indented 2 spaces
+```
 
 **Issue**: Agent is not using the Grounding Service tool