answering: dedupe sources by URL and add breadcrumb titles for course docs

- resolve_sources collapsed duplicates by chunk id, so multiple cited chunks of
  the same page (same URL) appeared as repeated source lines. Dedupe by URL.
- Course doc sources now render as a breadcrumb (Courses > LLM Zoomcamp > Project)
  so the reader can place the page in the course nav.

Author: alexeygrigorev

src/faq_assistant/answering.py

@@ -269,20 +269,38 @@ def resolve_sources(config, rag_answer: RagAnswer, results: list[SearchResult])
     sources: list[dict] = []
     for source_id in rag_answer.source_ids:
         result = by_id.get(source_id)
-        if result is None or result.id in seen:
+        if result is None:
             continue
-        seen.add(result.id)
+        # Collapse multiple cited chunks of the same page/entry: they share a URL,
+        # so without this the same doc shows up several times in the source list.
+        dedup_key = result.url or result.id
+        if dedup_key in seen:
+            continue
+        seen.add(dedup_key)
         sources.append(
             {
                 "id": result.id,
                 "source": SOURCE_LABELS.get(result.source_type, result.source_type),
-                "title": result.title,
+                "title": _source_title(config, result),
                 "url": result.url,
             }
         )
     return sources[: int(config["answering"]["max_sources"])]
 
 
+def _source_title(config, result: SearchResult) -> str:
+    """Display title for a cited source.
+
+    Course doc pages get a breadcrumb ("Courses > LLM Zoomcamp > Project") so the
+    reader can place the page in the course nav; other sources keep their own title.
+    """
+    if result.source_type == "course_docs":
+        course_name = config["courses"].get(result.course, {}).get("name") or result.course
+        parts = [part for part in ("Courses", course_name, result.title) if part]
+        return " > ".join(parts)
+    return result.title
+
+
 def build_context(results: list[SearchResult]) -> str:
     lines: list[str] = []
     for position, result in enumerate(results, start=1):

tests/test_answering.py

@@ -53,6 +53,27 @@ def test_resolve_sources_maps_labels_and_dedupes(cfg):
     ]
 
 
+def test_resolve_sources_collapses_chunks_sharing_a_url(cfg):
+    # Two different chunk ids of the same page share one URL -> one source line.
+    results = [
+        SearchResult(id="cd:1", source_type="course_docs", course="llm-zoomcamp", title="Project", url="u"),
+        SearchResult(id="cd:2", source_type="course_docs", course="llm-zoomcamp", title="Project", url="u"),
+    ]
+    rag = RagAnswer(answer="a", found_answer=True, source_ids=["cd:1", "cd:2"])
+    out = resolve_sources(cfg, rag, results)
+    assert len(out) == 1
+    assert out[0]["url"] == "u"
+
+
+def test_resolve_sources_course_docs_get_breadcrumb_title(cfg):
+    results = [
+        SearchResult(id="cd:1", source_type="course_docs", course="llm-zoomcamp", title="Project", url="u3"),
+    ]
+    rag = RagAnswer(answer="a", found_answer=True, source_ids=["cd:1"])
+    out = resolve_sources(cfg, rag, results)
+    assert out[0]["title"] == "Courses > LLM Zoomcamp > Project"
+
+
 def test_resolve_sources_empty_when_not_found(cfg):
     rag = RagAnswer(answer="no", found_answer=False, source_ids=["faq:1"])
     assert resolve_sources(cfg, rag, _results()) == []