Spaces:
Paused
Paused
github-actions[bot] commited on
Commit ·
58726c2
1
Parent(s): c6acc20
Deploy from GitHub: e41cf0cdffa82676b7c2034ffdca57d930b5e475
Browse files- app/routers/disease.py +14 -3
- app/services/llm_service.py +45 -6
app/routers/disease.py
CHANGED
|
@@ -17,7 +17,10 @@ def _looks_agri_product(title: str | None, source: str | None) -> bool:
|
|
| 17 |
text = f"{title or ''} {source or ''}".lower()
|
| 18 |
agri_keywords = [
|
| 19 |
"fungicide", "pesticide", "insecticide", "herbicide", "fertilizer",
|
| 20 |
-
"agri", "agriculture", "crop", "plant", "seed", "biofungicide"
|
|
|
|
|
|
|
|
|
|
| 21 |
]
|
| 22 |
bad_keywords = [
|
| 23 |
"amplifier", "speaker", "audio", "guitar", "home", "quilt", "comforter"
|
|
@@ -38,11 +41,19 @@ async def process_disease(request: DiseaseRequest):
|
|
| 38 |
seen_links = set()
|
| 39 |
for item in extracted_products:
|
| 40 |
product_name = (item.get("name") or "").strip() if isinstance(item, dict) else ""
|
|
|
|
| 41 |
if not product_name:
|
| 42 |
continue
|
| 43 |
|
| 44 |
-
#
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
search_results = search_shopping(search_query)
|
| 47 |
shopping = search_results.get('shopping', [])
|
| 48 |
|
|
|
|
| 17 |
text = f"{title or ''} {source or ''}".lower()
|
| 18 |
agri_keywords = [
|
| 19 |
"fungicide", "pesticide", "insecticide", "herbicide", "fertilizer",
|
| 20 |
+
"agri", "agriculture", "crop", "plant", "seed", "biofungicide",
|
| 21 |
+
"garden", "horticulture", "foliar", "micronutrient", "nutrient",
|
| 22 |
+
"calcium", "magnesium", "zinc", "boron", "sulfate", "sulphate",
|
| 23 |
+
"npk", "chelate", "soil conditioner", "plant tonic"
|
| 24 |
]
|
| 25 |
bad_keywords = [
|
| 26 |
"amplifier", "speaker", "audio", "guitar", "home", "quilt", "comforter"
|
|
|
|
| 41 |
seen_links = set()
|
| 42 |
for item in extracted_products:
|
| 43 |
product_name = (item.get("name") or "").strip() if isinstance(item, dict) else ""
|
| 44 |
+
category = (item.get("category") or "other").strip().lower() if isinstance(item, dict) else "other"
|
| 45 |
if not product_name:
|
| 46 |
continue
|
| 47 |
|
| 48 |
+
# Use category-aware query; previous fungicide-only bias caused misses for nutrient disorders.
|
| 49 |
+
if category == "nutrient":
|
| 50 |
+
suffix = "agriculture plant nutrient foliar spray india"
|
| 51 |
+
elif category in ("fungicide", "pesticide", "bio"):
|
| 52 |
+
suffix = f"agriculture {category} crop treatment india"
|
| 53 |
+
else:
|
| 54 |
+
suffix = "agriculture crop plant treatment india"
|
| 55 |
+
|
| 56 |
+
search_query = f"{product_name} {suffix}"
|
| 57 |
search_results = search_shopping(search_query)
|
| 58 |
shopping = search_results.get('shopping', [])
|
| 59 |
|
app/services/llm_service.py
CHANGED
|
@@ -1,5 +1,6 @@
|
|
| 1 |
import json
|
| 2 |
import base64
|
|
|
|
| 3 |
from app.config import settings
|
| 4 |
|
| 5 |
try:
|
|
@@ -76,11 +77,45 @@ def detect_disease(image_data: str, query: str = ""):
|
|
| 76 |
|
| 77 |
|
| 78 |
def extract_agri_products(detection_text: str):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
prompt = (
|
| 80 |
-
"From the disease analysis below, extract only AGRICULTURAL
|
| 81 |
-
"
|
|
|
|
| 82 |
"Return STRICT JSON only with this format: "
|
| 83 |
-
"{\"products\": [{\"name\": \"...\", \"category\": \"fungicide|pesticide|bio|other\"}]}. "
|
| 84 |
"Do not include non-agriculture items. If none are present return {\"products\": []}. "
|
| 85 |
f"Analysis: {detection_text}"
|
| 86 |
)
|
|
@@ -88,7 +123,8 @@ def extract_agri_products(detection_text: str):
|
|
| 88 |
oai = _get_openai_client()
|
| 89 |
completion = oai.chat.completions.create(
|
| 90 |
model="gpt-4o-mini",
|
| 91 |
-
temperature=0
|
|
|
|
| 92 |
messages=[{"role": "user", "content": prompt}],
|
| 93 |
)
|
| 94 |
extraction_text = completion.choices[0].message.content or ""
|
|
@@ -96,9 +132,12 @@ def extract_agri_products(detection_text: str):
|
|
| 96 |
extraction_text = _clean_fenced_json(extraction_text)
|
| 97 |
try:
|
| 98 |
parsed = json.loads(extraction_text)
|
| 99 |
-
|
|
|
|
|
|
|
|
|
|
| 100 |
except json.JSONDecodeError:
|
| 101 |
-
return
|
| 102 |
|
| 103 |
def chat_bot(query: str, image_data: bytes = None):
|
| 104 |
system_instruction = (
|
|
|
|
| 1 |
import json
|
| 2 |
import base64
|
| 3 |
+
import re
|
| 4 |
from app.config import settings
|
| 5 |
|
| 6 |
try:
|
|
|
|
| 77 |
|
| 78 |
|
| 79 |
def extract_agri_products(detection_text: str):
|
| 80 |
+
def _heuristic_product_candidates(text: str):
|
| 81 |
+
# Fallback parser from chemical solution sections when model extraction is empty/invalid.
|
| 82 |
+
candidates = []
|
| 83 |
+
seen = set()
|
| 84 |
+
|
| 85 |
+
patterns = [
|
| 86 |
+
r"active\s+ingredient\(s\)\s*:\s*(.+)",
|
| 87 |
+
r"product\s+type\s*:\s*(.+)",
|
| 88 |
+
r"recommended\s+products?\s*:\s*(.+)",
|
| 89 |
+
]
|
| 90 |
+
|
| 91 |
+
for line in (text or "").splitlines():
|
| 92 |
+
clean = line.strip(" -•*\t")
|
| 93 |
+
if not clean:
|
| 94 |
+
continue
|
| 95 |
+
|
| 96 |
+
for p in patterns:
|
| 97 |
+
m = re.search(p, clean, flags=re.IGNORECASE)
|
| 98 |
+
if not m:
|
| 99 |
+
continue
|
| 100 |
+
values = re.split(r",|/|\band\b", m.group(1), flags=re.IGNORECASE)
|
| 101 |
+
for value in values:
|
| 102 |
+
name = value.strip(" .:;-")
|
| 103 |
+
if len(name) < 3:
|
| 104 |
+
continue
|
| 105 |
+
key = name.lower()
|
| 106 |
+
if key in seen:
|
| 107 |
+
continue
|
| 108 |
+
seen.add(key)
|
| 109 |
+
candidates.append({"name": name, "category": "other"})
|
| 110 |
+
|
| 111 |
+
return candidates[:8]
|
| 112 |
+
|
| 113 |
prompt = (
|
| 114 |
+
"From the disease analysis below, extract only AGRICULTURAL treatment inputs that a farmer can buy "
|
| 115 |
+
"(fungicides, pesticides, bio-products, micronutrients, plant supplements, soil amendments, etc.). "
|
| 116 |
+
"Prefer active ingredients and practical market search terms. "
|
| 117 |
"Return STRICT JSON only with this format: "
|
| 118 |
+
"{\"products\": [{\"name\": \"...\", \"category\": \"fungicide|pesticide|bio|nutrient|other\"}]}. "
|
| 119 |
"Do not include non-agriculture items. If none are present return {\"products\": []}. "
|
| 120 |
f"Analysis: {detection_text}"
|
| 121 |
)
|
|
|
|
| 123 |
oai = _get_openai_client()
|
| 124 |
completion = oai.chat.completions.create(
|
| 125 |
model="gpt-4o-mini",
|
| 126 |
+
temperature=0,
|
| 127 |
+
response_format={"type": "json_object"},
|
| 128 |
messages=[{"role": "user", "content": prompt}],
|
| 129 |
)
|
| 130 |
extraction_text = completion.choices[0].message.content or ""
|
|
|
|
| 132 |
extraction_text = _clean_fenced_json(extraction_text)
|
| 133 |
try:
|
| 134 |
parsed = json.loads(extraction_text)
|
| 135 |
+
products = parsed.get("products", []) if isinstance(parsed, dict) else []
|
| 136 |
+
if isinstance(products, list) and products:
|
| 137 |
+
return products
|
| 138 |
+
return _heuristic_product_candidates(detection_text)
|
| 139 |
except json.JSONDecodeError:
|
| 140 |
+
return _heuristic_product_candidates(detection_text)
|
| 141 |
|
| 142 |
def chat_bot(query: str, image_data: bytes = None):
|
| 143 |
system_instruction = (
|