fix: bound dashboard embedding text

Sync GitHub commit 29b7083; bound README text used for retrieval embeddings while keeping full README/app source for MiniCPM quest analysis.

data/project_index.json

@@ -1,7 +1,7 @@
 {
   "schema_version": 3,
   "algorithm": "llama-cpp-embedding-v1",
-  "generated_at": "2026-06-07T11:52:16+00:00",
+  "generated_at": "2026-06-08T00:28:50+00:00",
   "snapshot_generated_at": "2026-06-07T11:51:09+00:00",
   "snapshot_source": "https://huggingface.co/api/spaces?author=build-small-hackathon",
   "snapshot_digest": "a442d8146fa01965567a27e43d42587de7048b93352e016151a524d2ecbf2034",
@@ -14,7 +14,7 @@
     "normalized": true,
     "build_source": "local",
     "builder": "scripts/build_project_index.py",
-    "llama_cpp_python_version": "0.3.26",
+    "llama_cpp_python_version": "0.3.27",
     "n_ctx": 2048
   },
   "documents": [

hackathon_advisor/data.py

@@ -32,6 +32,7 @@ DEFAULT_EMBEDDING_MODEL_REPO = "ggml-org/embeddinggemma-300m-qat-q8_0-GGUF"
 DEFAULT_EMBEDDING_MODEL_FILE = "embeddinggemma-300m-qat-Q8_0.gguf"
 DEFAULT_EMBEDDING_RUNTIME = "llama.cpp via llama-cpp-python"
 APP_FILE_EMBEDDING_CHAR_LIMIT = 8000
+README_EMBEDDING_CHAR_LIMIT = 4000
 
 
 EmbeddingFunction = Callable[[str], Sequence[float]]
@@ -91,7 +92,9 @@ class Project:
                 f"title: {self.title}",
                 f"slug: {self.slug.replace('-', ' ').replace('_', ' ')}",
                 f"summary: {self.summary}",
-                f"readme:\n{self.readme_body}" if self.readme_body else "",
+                f"readme:\n{bounded_embedding_text(self.readme_body, README_EMBEDDING_CHAR_LIMIT)}"
+                if self.readme_body
+                else "",
                 f"tags: {' '.join(self.tags)}",
                 f"models: {' '.join(self.models)}",
                 f"datasets: {' '.join(self.datasets)}",

tests/test_data.py

@@ -85,6 +85,30 @@ def test_searchable_text_includes_main_app_file_signals() -> None:
     assert "Project idea" in searchable
 
 
+def test_searchable_text_bounds_readme_body_for_embedding() -> None:
+    project = Project(
+        id="build-small-hackathon/long-readme",
+        title="Long README",
+        summary="",
+        tags=(),
+        models=(),
+        datasets=(),
+        likes=0,
+        sdk="gradio",
+        license="",
+        created_at="",
+        last_modified="",
+        host="",
+        url="https://example.test",
+        readme_body="a" * 2500 + "middle should not be embedded" + "b" * 2500,
+    )
+
+    searchable = project.searchable_text
+
+    assert "middle should not be embedded" not in searchable
+    assert len(searchable) < 4300
+
+
 def test_project_index_rejects_mismatched_snapshot(tmp_path: Path) -> None:
     payload = json.loads(Path("data/project_index.json").read_text(encoding="utf-8"))
     payload["snapshot_generated_at"] = "2000-01-01T00:00:00+00:00"