From 62dc69374a430fc686906916bcf746ae73317368 Mon Sep 17 00:00:00 2001
From: Philipp Mock <p.mock@iwm-tuebingen.de>
Date: Tue, 3 Feb 2026 11:44:11 +0100
Subject: [PATCH] added logging of RAG chunks for debugging

---
 local_rag.py        | 39 ++++++++++++++++++++++++---------------
 server.py           | 24 ++++++++++++++++++++++--
 templates/chat.html |  9 +++++++++
 3 files changed, 55 insertions(+), 17 deletions(-)

diff --git a/local_rag.py b/local_rag.py
index 70ded93..403b040 100644
--- a/local_rag.py
+++ b/local_rag.py
@@ -134,18 +134,26 @@ class LocalRAG:
         return doc_list
     
     def query(self, question, k=4):
-        """Query the RAG system"""
+        """Query the RAG system. Returns dict with 'answer' and 'retrieved' (list of chunks with content, source, page)."""
         if self.vectorstore is None:
-            return "Error: No documents loaded. Please add documents first."
-        
-        print(f"\nSearching for relevant documents...")
+            return {
+                "answer": "Error: No documents loaded. Please add documents first.",
+                "retrieved": [],
+            }
+
         docs = self.vectorstore.similarity_search(question, k=k)
-        print(f"Found {len(docs)} relevant documents")
-        
+        retrieved = [
+            {
+                "content": doc.page_content,
+                "source": doc.metadata.get("source", ""),
+                "page": doc.metadata.get("page"),
+            }
+            for doc in docs
+        ]
+
         # Combine context from documents
         context = "\n\n".join([doc.page_content for doc in docs])
-        
-        # Create prompt
+
         prompt = f"""Use the following context to answer the question.
 If you don't know the answer, say that you don't know instead of making up an answer.
 
@@ -155,12 +163,11 @@ Context:
 Question: {question}
 
 Answer:"""
-        
-        print("Generating answer with Ollama...")
+
         response = self.llm.invoke(prompt)
-        answer = response.content if hasattr(response, 'content') else str(response)
-        
-        return answer
+        answer = response.content if hasattr(response, "content") else str(response)
+
+        return {"answer": answer, "retrieved": retrieved}
 
 
 def main():
@@ -184,9 +191,11 @@ def main():
     
     # Query
     question = "What do the documents say about modality for perceived message perception?"
-    answer = rag.query(question)
+    result = rag.query(question)
     print(f"\nQuestion: {question}")
-    print(f"Answer: {answer}")
+    print(f"Answer: {result['answer']}")
+    if result.get("retrieved"):
+        print(f"Retrieved {len(result['retrieved'])} chunks")
     
     # print("\nSetup complete! Uncomment the code above to add documents and query.")
 
diff --git a/server.py b/server.py
index 48c7839..54e71a8 100644
--- a/server.py
+++ b/server.py
@@ -22,9 +22,16 @@ class ChatRequest(BaseModel):
     message: str
 
 
+class RetrievedChunk(BaseModel):
+    content: str
+    source: str
+    page: int | None
+
+
 class ChatResponse(BaseModel):
     answer: str
     error: str | None = None
+    retrieved: list[RetrievedChunk] | None = None
 
 
 @app.get("/", response_class=HTMLResponse)
@@ -42,8 +49,21 @@ def chat(request: ChatRequest):
     if not request.message or not request.message.strip():
         return ChatResponse(answer="", error="Message cannot be empty")
     try:
-        answer = rag.query(request.message.strip())
-        return ChatResponse(answer=answer)
+        result = rag.query(request.message.strip())
+        answer = result["answer"]
+        retrieved = result.get("retrieved", [])
+
+        # Server-side console trace: log retrieved chunks before LLM answer
+        if retrieved:
+            print(f"\n[RAG] Retrieved {len(retrieved)} chunk(s) for query: {request.message[:80]!r}")
+            for i, chunk in enumerate(retrieved):
+                content = chunk.get("content", "")
+                preview = (content[:1000] + "...") if len(content) > 1000 else content
+                print(f"  [{i + 1}] source={chunk.get('source', '')} page={chunk.get('page')} | {preview!r}")
+        else:
+            print(f"\n[RAG] Retrieved 0 chunks for query: {request.message[:80]!r}")
+
+        return ChatResponse(answer=answer, retrieved=retrieved)
     except Exception as e:
         return ChatResponse(answer="", error=str(e))
 
diff --git a/templates/chat.html b/templates/chat.html
index 5588efa..fa414cb 100644
--- a/templates/chat.html
+++ b/templates/chat.html
@@ -186,6 +186,15 @@
         });
         const data = await res.json();
         setLoading(false);
+        if (data.retrieved && data.retrieved.length > 0) {
+          console.groupCollapsed('[RAG] Retrieved ' + data.retrieved.length + ' chunk(s)');
+          data.retrieved.forEach(function(chunk, i) {
+            console.group('Chunk ' + (i + 1) + ' | ' + (chunk.source || '') + (chunk.page != null ? ' p.' + chunk.page : ''));
+            console.log(chunk.content);
+            console.groupEnd();
+          });
+          console.groupEnd();
+        }
         if (data.error) {
           appendMessage('assistant', data.error, true);
         } else {