docs: update repository and scripts readmes to reflect 4.2GB master sync layouts

README.md

@@ -46,14 +46,17 @@ os.makedirs("/content/SafeVixAI/chatbot_service", exist_ok=True)
 
 ```
 SafeVixAI-Dataset-Hub/
-├── data/                         ← 3.6 GB of raw intelligence data
-│   ├── chatbot_service/data/     ← Legal PDFs, GIS CSVs, accident data, models
-│   └── backend/datasets/         ← Challan rules, road infrastructure
+├── data/                         ← 4.2 GB of raw intelligence data (11,008 files)
+│   ├── backend/
+│   │   ├── data/                 ← Civic intel (OSM data, toll plazas, ward boundaries)
+│   │   └── datasets/             ← Raw database assets (blackspots, violations CSVs)
+│   ├── chatbot_service/data/     ← Chatbot reference directories & built vector stores
+│   └── frontend/offline-data/    ← Regional translation matrices & offline PWA bundles
 │
-└── scripts/                      ← Reproducible data acquisition pipeline
-    ├── scripts/data/             ← Root-level data scripts (fetchers, extractors)
-    ├── backend/data/             ← Backend data transforms (pure Python)
-    └── chatbot_service/data/     ← Pro Overpass GIS fetchers
+└── scripts/                      ← Complete data acquisition & processing pipelines
+    ├── backend/                  ← Core backend and database migration scripts
+    ├── chatbot_service/          ← Chatbot agent and QA validation scripts
+    └── scripts/                  ← Legacy scrapers, downloaders, and seeders
 ```
 
 ---
@@ -68,6 +71,7 @@ SafeVixAI-Dataset-Hub/
 | Road Infrastructure | `pmgsy_roads.geojson`, `toll_plazas.csv` | ~900 MB | PMGSY GeoSadak / NHAI |
 | Hospital Directory | `hospital_directory.csv`, `nin_facilities.csv` | ~1.2 GB | NHP / NIN |
 | Traffic Violations | `violations_seed.csv`, `state_overrides.csv` | ~5 MB | MVA 2019 |
+| PWA Translations | `translations/*.json` (11 regional Indian languages) | ~6.4 MB | Automated DeepL/Google Sync |
 | Road Damage Model | `road_damage_2025/` | ~800 MB | ONNX + Training Data |
 | QA Pairs | `qa_pairs/` | ~50 MB | Custom RAG Training |
 

data/chatbot_service/data/qa_pairs/mhqa-main/benchmarking/bert_evaluate.py

@@ -49,7 +49,8 @@ def create_dataloader(dataset, tokenizer, batch_size, max_len):
                                 batch_size=batch_size,
                                 sampler=eval_sampler,
                                 collate_fn=model_collate_fn, 
-                                num_workers=1)
+                                num_workers=1,
+                                pin_memory=True)
     return test_dataloader 
 
 def run_inference(model, dataloader):

data/chatbot_service/data/qa_pairs/mhqa-main/benchmarking/model.py

@@ -36,7 +36,10 @@ class OpenAIGPT(BaseModel):
             max_tokens=1024,
             stream=False,
             temperature=0.7,
+            user="mhqa_benchmarking",
         )
+        if getattr(response.choices[0].message, "refusal", None):
+            raise ValueError(f"OpenAI Refusal: {response.choices[0].message.refusal}")
         generated_text = response.choices[0].message.content
         print(generated_text)
         return generated_text
@@ -73,7 +76,10 @@ class GroqModel(BaseModel):
             ],
             max_tokens=1024,
             temperature=0.7,
+            user="mhqa_benchmarking",
         )
+        if getattr(response.choices[0].message, "refusal", None):
+            raise ValueError(f"Groq Refusal: {response.choices[0].message.refusal}")
         return response.choices[0].message.content
 
 class HFApiModel(BaseModel):

data/chatbot_service/data/qa_pairs/mhqa-main/difficulty_evaluation/model.py

@@ -36,7 +36,10 @@ class OpenAIGPT(BaseModel):
             max_tokens=1024,
             stream=False,
             temperature=0.7,
+            user="mhqa_difficulty_evaluation",
         )
+        if getattr(response.choices[0].message, "refusal", None):
+            raise ValueError(f"OpenAI Refusal: {response.choices[0].message.refusal}")
         generated_text = response.choices[0].message.content
         print(generated_text)
         return generated_text
@@ -73,7 +76,10 @@ class GroqModel(BaseModel):
             ],
             max_tokens=1024,
             temperature=0.7,
+            user="mhqa_difficulty_evaluation",
         )
+        if getattr(response.choices[0].message, "refusal", None):
+            raise ValueError(f"Groq Refusal: {response.choices[0].message.refusal}")
         return response.choices[0].message.content
 
 class HFApiModel(BaseModel):

data/chatbot_service/data/qa_pairs/mhqa-main/question_types/model.py

@@ -37,7 +37,10 @@ class OpenAIGPT(BaseModel):
             stream=False,
             temperature=0.7,
             seed=42,
+            user="mhqa_question_types",
         )
+        if getattr(response.choices[0].message, "refusal", None):
+            raise ValueError(f"OpenAI Refusal: {response.choices[0].message.refusal}")
         generated_text = response.choices[0].message.content
         print(generated_text)
         return generated_text
@@ -74,7 +77,10 @@ class GroqModel(BaseModel):
             ],
             max_tokens=1024,
             temperature=0.7,
+            user="mhqa_question_types",
         )
+        if getattr(response.choices[0].message, "refusal", None):
+            raise ValueError(f"Groq Refusal: {response.choices[0].message.refusal}")
         return response.choices[0].message.content
 
 class HFApiModel(BaseModel):

data/chatbot_service/data/qa_pairs/mhqa-main/sft/evaluate.py

@@ -35,7 +35,8 @@ def get_dataloader(dataset, tokenizer, args):
                                 batch_size=args['batch_size'],
                                 sampler=eval_sampler,
                                 collate_fn=model_collate_fn, 
-                                num_workers=95)
+                                num_workers=95,
+                                pin_memory=True)
     return test_dataloader
 
 def run_inference(model, dataloader, args):

data/chatbot_service/data/qa_pairs/mhqa-main/sft/model.py

@@ -124,7 +124,8 @@ class MCQAModel(pl.LightningModule):
                                 batch_size=self.batch_size,
                                 sampler=train_sampler,
                                 collate_fn=model_collate_fn,
-                                num_workers=95)
+                                num_workers=95,
+                                pin_memory=True)
     return train_dataloader
   
   def val_dataloader(self):
@@ -138,7 +139,8 @@ class MCQAModel(pl.LightningModule):
                                 batch_size=self.batch_size,
                                 sampler=eval_sampler,
                                 collate_fn=model_collate_fn,
-                                num_workers=95)
+                                num_workers=95,
+                                pin_memory=True)
     return val_dataloader
   
   def test_dataloader(self):
@@ -152,5 +154,6 @@ class MCQAModel(pl.LightningModule):
                                 batch_size=self.batch_size,
                                 sampler=eval_sampler,
                                 collate_fn=model_collate_fn, 
-                                num_workers=95)
+                                num_workers=95,
+                                pin_memory=True)
     return test_dataloader

scripts/README.md

@@ -1,6 +1,6 @@
 # SafeVixAI — Data Acquisition Scripts 🔬
 
-These scripts are the **raw data pipeline** that built the 3.6GB SafeVixAI dataset.
+These scripts are the **raw data pipeline** that built the 4.2 GB SafeVixAI dataset.
 All scripts here are **pure Python** — they require no database, no Redis, no PostGIS.
 
 > Scripts are mirrored from the [SafeVixAI main repo](https://github.com/SafeVixAI/SafeVixAI) and organized by their origin folder.
@@ -11,28 +11,18 @@ All scripts here are **pure Python** — they require no database, no Redis, no
 
 ```
 scripts/
-├── scripts/data/             ← from SafeVixAI/scripts/data/
-│   ├── fetch_*.py            ← Overpass GIS fetchers (basic version)
-│   ├── _overpass_utils.py    ← Core GIS utility
-│   ├── download_legal_pdfs.py
-│   ├── extract_morth2022_tables.py
-│   ├── seed_blackspots.py
-│   ├── bootstrap_local_data.py
-│   ├── verify_data.py
-│   ├── inspect_zips.py
-│   ├── audit_env.py
-│   ├── check_all_scripts.py
-│   └── setup_kaggle.ps1
+├── backend/                  ← from SafeVixAI/backend/scripts/
+│   ├── app/                  ← Web app database setup & SQL migrations
+│   └── data/                 ← Civic intel ETL pipelines & seeders
 │
-├── backend/data/             ← from SafeVixAI/backend/scripts/data/
-│   ├── seed_violations.py    ← MVA 2019 traffic fine normalizer
-│   ├── prepare_road_sources.py
-│   ├── sample_pmgsy.py
-│   └── road_sources.example.json
+├── chatbot_service/          ← from SafeVixAI/chatbot_service/scripts/
+│   ├── app/                  ← Chatbot seeders
+│   ├── data/                 ← Overpass GIS fetchers (Pro version)
+│   └── verify_rag.py         ← Vectorstore validation
 │
-└── chatbot_service/data/     ← from SafeVixAI/chatbot_service/scripts/data/
-    ├── _overpass_utils.py    ← ⭐ Pro version (retries + backoff)
-    └── fetch_*.py            ← ⭐ Pro fetchers (use these over scripts/data/)
+└── scripts/                  ← from SafeVixAI/scripts/
+    ├── app/                  ← Asset generators
+    └── data/                 ← Scrapers, downloaders, and legacy seeders
 ```
 
 ---

scripts/backend/data/fetch_osm_civic_features.py

@@ -123,7 +123,12 @@ def save_features(city: str, features: dict[str, list[dict]]) -> dict[str, int]:
         if not items:
             continue
 
-        outfile = OUTPUT_DIR / f'{city}_{ftype}.csv'
+        # Resolve and validate the target file path to prevent path traversal
+        resolved_output_dir = os.path.realpath(str(OUTPUT_DIR))
+        outfile = os.path.realpath(str(OUTPUT_DIR / f'{city}_{ftype}.csv'))
+        if not outfile.startswith(resolved_output_dir + os.sep):
+            raise ValueError(f"Path traversal detected: {outfile} is outside of {resolved_output_dir}")
+
         fieldnames = ['osm_id', 'lat', 'lon', 'feature_type', 'city', 'tags']
         with open(outfile, 'w', newline='', encoding='utf-8') as f:
             writer = csv.DictWriter(f, fieldnames=fieldnames)