fix: reduce embedding context

Sync GitHub commit 663bcb5; use n_ctx=768 for dashboard retrieval embeddings.

README.md

@@ -223,6 +223,7 @@ ADVISOR_CACHE_DIR=/data/advisor-cache
 ADVISOR_REFRESH_EMBEDDING_TIMEOUT_SECONDS=1800
 ADVISOR_EMBEDDING_MODEL_REPO=ggml-org/embeddinggemma-300m-qat-q8_0-GGUF
 ADVISOR_EMBEDDING_MODEL_FILE=embeddinggemma-300m-qat-Q8_0.gguf
+ADVISOR_EMBEDDING_N_CTX=768
 ADVISOR_ASR_MODEL_ID=nvidia/nemotron-speech-streaming-en-0.6b
 ```
 

data/project_index.json

@@ -1,7 +1,7 @@
 {
   "schema_version": 3,
   "algorithm": "llama-cpp-embedding-v1",
-  "generated_at": "2026-06-08T00:34:45+00:00",
+  "generated_at": "2026-06-08T00:39:11+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",
@@ -15,7 +15,7 @@
     "build_source": "local",
     "builder": "scripts/build_project_index.py",
     "llama_cpp_python_version": "0.3.27",
-    "n_ctx": 2048
+    "n_ctx": 768
   },
   "documents": [
     {

hackathon_advisor/llama_embedding.py

@@ -19,7 +19,7 @@ from hackathon_advisor.data import (
 
 TRUE_VALUES = {"1", "true", "yes", "on"}
 FALSE_VALUES = {"0", "false", "no", "off"}
-DEFAULT_N_CTX = 2048
+DEFAULT_N_CTX = 768
 
 
 class LlamaCppEmbedder: