From 33b960f786ee459021349f5a8d3d4b3c58b820ac Mon Sep 17 00:00:00 2001
From: Woody <woody.ck.tse@gmail.com>
Date: Thu, 23 Apr 2026 16:28:07 +0800
Subject: [PATCH] fix(backend): extract JSON from markdown code blocks in LLM
 responses

The LLM (Qwen3.5 via OpenRouter) returns JSON wrapped in markdown code blocks:

```json

["project manager", "limits", ...]

```

But the code was trying to parse this directly with json.loads(), causing:

- QueryDecomposer to return empty keywords

- RelevanceFilter to fail with "Expecting value: line 1 column 1"

Changes:

- Added _extract_json_from_markdown() helper function to both modules

- Strips markdown code block markers (```json and ```) before JSON parsing

- Added unit tests for markdown code block handling

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus \u003cclio-agent@sisyphuslabs.ai\u003e
---
 backend/app/services/query_decomposer.py         | 13 +++++++++++++
 backend/app/services/relevance_filter.py         | 12 ++++++++++++
 backend/app/test/test_phase1_query_decomposer.py | 14 ++++++++++++++
 backend/app/test/test_phase1_relevance_filter.py | 12 ++++++++++++
 4 files changed, 51 insertions(+)

diff --git a/backend/app/services/query_decomposer.py b/backend/app/services/query_decomposer.py
index 42a63dc..3950ddc 100644
--- a/backend/app/services/query_decomposer.py
+++ b/backend/app/services/query_decomposer.py
@@ -8,12 +8,23 @@ from __future__ import annotations
 
 import json
 import logging
+import re
 from typing import List
 
 
 logger = logging.getLogger(__name__)
 
 
+def _extract_json_from_markdown(response: str) -> str:
+    if not isinstance(response, str):
+        return str(response)
+    pattern = r"```(?:json)?\s*\n?(.*?)\n?```"
+    match = re.search(pattern, response, re.DOTALL)
+    if match:
+        return match.group(1).strip()
+    return response.strip()
+
+
 class QueryDecomposer:
     """Decompose a natural language question into a list of keywords.
 
@@ -51,6 +62,8 @@ class QueryDecomposer:
         if not isinstance(response, str):
             response = str(response)
 
+        response = _extract_json_from_markdown(response)
+
         try:
             data = json.loads(response)
         except json.JSONDecodeError:
diff --git a/backend/app/services/relevance_filter.py b/backend/app/services/relevance_filter.py
index e5d933c..c783f0f 100644
--- a/backend/app/services/relevance_filter.py
+++ b/backend/app/services/relevance_filter.py
@@ -2,12 +2,23 @@ from __future__ import annotations
 
 import json
 import logging
+import re
 from typing import List, Tuple, Dict
 
 
 logger = logging.getLogger(__name__)
 
 
+def _extract_json_from_markdown(response: str) -> str:
+    if not isinstance(response, str):
+        return str(response)
+    pattern = r"```(?:json)?\s*\n?(.*?)\n?```"
+    match = re.search(pattern, response, re.DOTALL)
+    if match:
+        return match.group(1).strip()
+    return response.strip()
+
+
 class RelevanceFilter:
     """RelevanceFilter batches chunk texts to an LLM and selects those with
     relevance scores above a threshold.
@@ -43,6 +54,7 @@ class RelevanceFilter:
 
         scores: List[float] = []
         try:
+            response = _extract_json_from_markdown(response)
             parsed = json.loads(response)
             if not isinstance(parsed, list):
                 return []
diff --git a/backend/app/test/test_phase1_query_decomposer.py b/backend/app/test/test_phase1_query_decomposer.py
index 8d5e793..f57c9b0 100644
--- a/backend/app/test/test_phase1_query_decomposer.py
+++ b/backend/app/test/test_phase1_query_decomposer.py
@@ -55,3 +55,17 @@ async def test_decompose_mixed_types_coerced_to_strings():
     decomposer = QueryDecomposer(llm)
     result = await decomposer.decompose("Question?")
     assert result == ["a", "2", "None"]
+
+
+async def test_decompose_json_in_markdown_code_block():
+    llm = MockLLMClient('```json\n["project", "manager", "limits"]\n```')
+    decomposer = QueryDecomposer(llm)
+    result = await decomposer.decompose("What are the limits?")
+    assert result == ["project", "manager", "limits"]
+
+
+async def test_decompose_json_in_plain_code_block():
+    llm = MockLLMClient('```\n["alpha", "beta"]\n```')
+    decomposer = QueryDecomposer(llm)
+    result = await decomposer.decompose("Keywords?")
+    assert result == ["alpha", "beta"]
diff --git a/backend/app/test/test_phase1_relevance_filter.py b/backend/app/test/test_phase1_relevance_filter.py
index 93e158b..54e2bc4 100644
--- a/backend/app/test/test_phase1_relevance_filter.py
+++ b/backend/app/test/test_phase1_relevance_filter.py
@@ -66,3 +66,15 @@ async def test_filter_all_outside_threshold():
     rf = RelevanceFilter(llm)
     result = await rf.filter("Question", chunks, threshold=5.0)
     assert result == []
+
+
+async def test_filter_json_in_markdown_code_block():
+    chunks = _make_chunks()
+    llm = MagicMock()
+    llm.complete = AsyncMock(return_value="```json\n[8.0, 3.0, 9.0]\n```")
+
+    rf = RelevanceFilter(llm)
+    result = await rf.filter("Question", chunks, threshold=7.0)
+
+    expected = [chunks[0], chunks[2]]
+    assert result == expected