diff --git a/agent/indexing/vector_store.py b/agent/indexing/vector_store.py
index 41b615a..d180095 100644
--- a/agent/indexing/vector_store.py
+++ b/agent/indexing/vector_store.py
@@ -1,7 +1,9 @@
 import chromadb
 
+# safer collection creation
 client = chromadb.Client()
-collection = client.create_collection(name="codebase")
+collection = client.get_or_create_collection(name="codebase")
+
 
 def store_embeddings(chunks, embeddings):
     for i, chunk in enumerate(chunks):
@@ -11,23 +13,48 @@ def store_embeddings(chunks, embeddings):
             ids=[str(i)]
         )
 
+
+MIN_CHUNK_LENGTH = 30  # Phase 4 threshold
+
+
 def query_embeddings(query_embedding, k=5):
     results = collection.query(
         query_embeddings=[query_embedding],
         n_results=k
     )
 
-    docs = results.get("documents", [[]])[0]
+    docs = results.get("documents", [[]])
+
+    # safety check
+    if not docs or not docs[0]:
+        print("No documents retrieved")
+        return []
+
+    docs = docs[0]
+
+    # 🧹 CLEAN: remove empty / noisy chunks
+    cleaned = [
+        doc for doc in docs
+        if doc and len(doc.strip()) > MIN_CHUNK_LENGTH
+    ]
+
+    # 🎯 PRIORITIZE: functions/classes
+    filtered = [
+        doc for doc in cleaned
+        if "def " in doc or "class " in doc
+    ]
+
+    # fallback if nothing found
+    if not filtered:
+        filtered = cleaned
 
-    # remove empty / tiny chunks
-    cleaned = [doc for doc in docs if doc and len(doc.strip()) > 20]
+    # 🔁 REMOVE DUPLICATES
+    unique_docs = list(dict.fromkeys(filtered))
 
-    # 🔥 prioritize useful code (functions/classes)
-    filtered = []
-    for doc in cleaned:
-        if "def " in doc or "class " in doc:
-            filtered.append(doc)
+    # 🧪 DEBUG LOGS
+    print(f"Retrieved {len(unique_docs)} relevant chunks")
 
-    print(f"Retrieved {len(filtered)} relevant chunks")
+    for i, doc in enumerate(unique_docs[:2]):
+        print(f"Chunk {i+1} preview:", doc[:100])
 
-    return filtered[:k]
\ No newline at end of file
+    return unique_docs[:k]
\ No newline at end of file
diff --git a/tests/test_generated.py b/tests/test_generated.py
index ff04abe..ee02b1e 100644
--- a/tests/test_generated.py
+++ b/tests/test_generated.py
@@ -1,26 +1,34 @@
 import pytest
-import os
-from agent.github.committer import commit_tests
-from agent.main import get_pr_diff
+from your_module import store_embeddings, query_embeddings
 
-def test_commit_tests():
-    commit_tests()
+def test_store_embeddings():
+    chunks = ["chunk1", "chunk2"]
+    embeddings = [[1, 2], [3, 4]]
+    store_embeddings(chunks, embeddings)
 
-def test_get_pr_diff_empty():
-    with pytest.raises(subprocess.CalledProcessError):
-        get_pr_diff()
+def test_query_embeddings():
+    query_embedding = [1, 2]
+    results = query_embeddings(query_embedding)
+    assert isinstance(results, list)
 
-def test_commit_tests_exception():
-    try:
-        commit_tests()
-    except Exception as e:
-        assert str(e)
+def test_query_embeddings_empty():
+    collection = client.get_or_create_collection(name="empty")
+    query_embedding = [1, 2]
+    results = collection.query(query_embeddings=[query_embedding], n_results=5)
+    assert results.get("documents", []) == [[]]
 
-def test_get_pr_diff_no_diff():
-    diff = get_pr_diff()
-    assert diff.strip() == ""
+def test_query_embeddings_min_length():
+    query_embedding = [1, 2]
+    chunks = ["a" * 29, "b" * 31]
+    embeddings = [[1, 2], [3, 4]]
+    store_embeddings(chunks, embeddings)
+    results = query_embeddings(query_embedding)
+    assert len(results) == 1
 
-def test_commit_tests_push():
-    subprocess.run(["git", "config", "user.name", "github-actions"], check=True)
-    subprocess.run(["git", "config", "user.email", "actions@github.com"], check=True)
-    commit_tests()
\ No newline at end of file
+def test_query_embeddings_k():
+    query_embedding = [1, 2]
+    chunks = ["a" * 31, "b" * 31, "c" * 31]
+    embeddings = [[1, 2], [3, 4], [5, 6]]
+    store_embeddings(chunks, embeddings)
+    results = query_embeddings(query_embedding, k=2)
+    assert len(results) == 2
\ No newline at end of file