initial commit

.gitignore

@@ -0,0 +1,13 @@
+../.env
+.venv/
+.deleted/
+test.ipynb
+test.py
+aliases/
+raft_state.json
+mlartifacts/
+mlruns/
+.deleted/
+secret.py
+docker-compose.yml
+.env

.idea/.gitignore

@@ -0,0 +1,5 @@
+# Default ignored files
+/shelf/
+/workspace.xml
+# Editor-based HTTP Client requests
+/httpRequests/

.idea/Legal-RAG-Chatbot.iml

@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<module type="PYTHON_MODULE" version="4">
+  <component name="NewModuleRootManager">
+    <content url="file://$MODULE_DIR$">
+      <excludeFolder url="file://$MODULE_DIR$/.venv" />
+    </content>
+    <orderEntry type="jdk" jdkName="Python 3.13 (Legal-RAG-Chatbot)" jdkType="Python SDK" />
+    <orderEntry type="sourceFolder" forTests="false" />
+  </component>
+</module>

.idea/inspectionProfiles/profiles_settings.xml

@@ -0,0 +1,6 @@
+<component name="InspectionProjectProfileManager">
+  <settings>
+    <option name="USE_PROJECT_PROFILE" value="false" />
+    <version value="1.0" />
+  </settings>
+</component>

.idea/misc.xml

@@ -0,0 +1,7 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="Black">
+    <option name="sdkName" value="Python 3.13 (Legal-RAG-Chatbot)" />
+  </component>
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.13 (Legal-RAG-Chatbot)" project-jdk-type="Python SDK" />
+</project>

.idea/modules.xml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectModuleManager">
+    <modules>
+      <module fileurl="file://$PROJECT_DIR$/.idea/Legal-RAG-Chatbot.iml" filepath="$PROJECT_DIR$/.idea/Legal-RAG-Chatbot.iml" />
+    </modules>
+  </component>
+</project>

.idea/vcs.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="VcsDirectoryMappings">
+    <mapping directory="" vcs="Git" />
+  </component>
+</project>

README.md

@@ -1,13 +1,14 @@
 ---
-title: Legal RAG Chatbot2
-emoji: 🏢
-colorFrom: blue
-colorTo: yellow
+title: Legal RAG Chatbot
+emoji: 💬
+colorFrom: yellow
+colorTo: purple
 sdk: gradio
-sdk_version: 5.42.0
+sdk_version: 5.0.1
 app_file: app.py
 pinned: false
-short_description: hello
+license: apache-2.0
+short_description: A RAG Chatbot that can answer legal questions
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).

app.py

@@ -0,0 +1,38 @@
+import gradio as gr
+from rag.rag_production import get_rag_chain
+
+def rag_fn(model_name: str, input: str):
+    try:
+        rag_chain = get_rag_chain(model_name=model_name)
+        response_stream = rag_chain.stream({'input': input})
+
+        full_answer = ''
+        for chunk in response_stream:
+            if 'answer' in chunk and chunk['answer'] is not None:
+                answer_piece = chunk['answer']
+                full_answer += answer_piece
+                yield full_answer
+    except Exception as e:
+        import traceback
+        print(traceback.format_exc())
+        yield f"An error occurred: {e}"
+
+interface = gr.Interface(
+    fn = rag_fn,
+    inputs = [
+      gr.Dropdown(choices=['llama-3.3-70b-versatile', 'openai/gpt-oss-120b'], label="MODEL"),
+      gr.Textbox(label='QUESTION'),
+    ],
+    outputs = gr.Textbox(label='ANSWER'),
+    title = "Legal RAG Chatbot",
+    description = "Select a model and ask a question to get an answer from the RAG system.",
+    examples = [
+['llama-3.3-70b-versatile', 'What is the maximum duration of determinate imprisonment that can be imposed on an offender?'],
+        ['openai/gpt-oss-120b','If someone voluntarily pays damages after committing a crime, how might this affect their sentencing?']
+    ],
+    cache_examples=False
+)
+
+if __name__ == "__main__":
+    interface.queue()
+    interface.launch()

data/processed_data/criminal_code_of_vietnam.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1eeb278744619c4a50994e9d04247094daeb1c2a0870eb54deba9657de4e6cb
+size 733443

embeddings/embedder.py

@@ -0,0 +1,102 @@
+import os
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_qdrant import QdrantVectorStore, RetrievalMode
+from qdrant_client import QdrantClient, models
+import logging
+import pickle
+from pathlib import Path
+
+logging.basicConfig(
+    level=logging.INFO,
+    format='%(asctime)s — %(levelname)s — %(message)s',
+)
+logger = logging.getLogger(__name__)
+
+def get_vectorstore() -> QdrantVectorStore:
+    base_dir = Path(__file__).resolve().parent.parent
+    doc_path = base_dir / 'data' / 'processed_data' / 'criminal_code_of_vietnam.pkl'
+
+    with open(doc_path, 'rb') as f:
+        doc_list = pickle.load(f)
+
+    qdrant_api_key = os.getenv('QDRANT_API_KEY')
+    qdrant_url = os.getenv('QDRANT_URL')
+    hf_api_key = os.getenv('HUGGINGFACEHUB_API_TOKEN')
+
+    collection_name = 'legal_db'
+    client = QdrantClient(url=qdrant_url, api_key=qdrant_api_key)
+
+    model_name = 'BAAI/bge-large-en'
+    model_kwargs = {'device': 'cpu'}
+    encode_kwargs = {'normalize_embeddings': False}
+
+    embeddings = HuggingFaceEmbeddings(
+        model_name=model_name,
+        model_kwargs=model_kwargs,
+        encode_kwargs=encode_kwargs
+    )
+    logger.info('Embedding created.')
+    dummy_embedding = embeddings.embed_query('A dummy to test embedding dimension')
+    vector_dim = len(dummy_embedding)
+
+    vectors_config = models.VectorParams(size=vector_dim, distance=models.Distance.COSINE)
+
+    if collection_name in [c.name for c in client.get_collections().collections]:
+        logger.info('Collection exists. Connecting...')
+        collection_info = client.get_collection(collection_name)
+        existing_dim = None
+        if hasattr(collection_info.config, 'vectors') and hasattr(collection_info.config.vectors, 'size'):
+            existing_dim = collection_info.config.vectors.size
+        elif hasattr(collection_info.config, 'params') and hasattr(collection_info.config.params, 'vectors') and hasattr(collection_info.config.params.vectors, 'size'):
+            existing_dim = collection_info.config.params.vectors.size
+
+        logger.info(f'Existing dimension: {existing_dim}')
+        if existing_dim != vector_dim:
+            raise ValueError(
+                f'Dimension mismatch: existing collection has {existing_dim}, but embedding model gives {vector_dim}'
+            )
+
+        db = QdrantVectorStore.from_existing_collection(
+            embedding=embeddings,
+            collection_name=collection_name,
+            prefer_grpc=False,
+            url=qdrant_url,
+            api_key = qdrant_api_key
+        )
+    else:
+        logger.info(f'Collection "{collection_name}" does not exist. Creating new collection...')
+        client.create_collection(
+            collection_name=collection_name,
+            vectors_config=vectors_config,
+
+        )
+        db = QdrantVectorStore.from_documents(
+            documents=doc_list,
+            embedding=embeddings,
+            url=qdrant_url,
+            prefer_grpc=False,
+            collection_name=collection_name,
+            retrieval_mode = RetrievalMode.DENSE,
+            api_key = qdrant_api_key
+        )
+        logger.info('Qdrant Index created.')
+
+        fields_to_index = {
+            'metadata.article': "keyword",
+            'metadata.chapter': "keyword",
+            'metadata.id': "keyword",
+            'metadata.source': "keyword",
+            'metadata.title': "keyword",
+        }
+
+        for field, schema in fields_to_index.items():
+            client.create_payload_index(
+                collection_name = collection_name,
+                field_name = field,
+                field_schema = schema,
+            )
+
+    return db
+
+
+

rag/rag_production.py

@@ -0,0 +1,91 @@
+from langchain.chains.combine_documents import create_stuff_documents_chain
+from langchain.chains.retrieval import create_retrieval_chain
+from langchain_groq import ChatGroq
+from langchain_core.prompts import ChatPromptTemplate
+import os
+from langchain_core.prompts import PromptTemplate
+import logging
+from embeddings.embedder import get_vectorstore
+
+
+logging.basicConfig(
+    level=logging.INFO,
+    format='%(asctime)s - %(name)s - %(levelName)s - %(message)s'
+)
+logger = logging.getLogger(__name__)
+
+
+rag_prompt_template = '''
+You are a legal assistant trained to answer questions using legal documents.
+
+If the answer cannot be determined from the available legal text, you must answer without include the
+According to Article <article>, Chapter <chapter>, <title> phrase, for example:
+   > The answer cannot be determined from the available legal text.
+
+Otherwise, you must STRICTLY follow this 3-step structure:
+
+1. **Begin your response ONLY with this format** (fill in the values from metadata):
+   > According to Article <article>, Chapter <chapter>, <title>:
+
+
+2. **Then, extract and summarize the most relevant point** from the provided context.
+
+3. **Finally, answer the user’s question clearly and formally**, referring only to the point above.
+
+---
+💥 IMPORTANT RULES:
+- Do NOT repeat or paraphrase the law reference later.
+- Do NOT invent any legal information — use ONLY the provided context and metadata.
+- Do NOT add phrase like 'The most relevant point is that' when mentioning the context.
+- If the context is insufficient, respond with:
+  > "The answer cannot be determined from the available legal text."
+---
+
+**User Question:**  
+{input}
+
+**Retrieved Legal Context:**  
+{context}
+'''
+
+document_prompt = PromptTemplate.from_template(
+    '''
+    Article: {article}
+    Chapter: {chapter}
+    Title: {title}
+    Content: {page_content}
+    '''
+)
+
+prompt = ChatPromptTemplate.from_messages([
+    ('system', rag_prompt_template),
+    ('user', "Context:\n{context}\n\nQuestion:\n{input}\n\nAnswer:")
+])
+
+def get_rag_chain(model_name='llama-3.3-70b-versatile', k=1):
+    db = get_vectorstore()
+
+    groq_api_key = os.getenv('GROQ_API_KEY')
+
+    llm = ChatGroq(
+    model_name=model_name,
+    temperature=0,
+    max_tokens=10000,
+    api_key=groq_api_key,
+)
+
+    retriever = db.as_retriever(
+        search_type = 'similarity',
+        search_kwargs = {'k': k}
+    )
+
+    combine_doc_chain = create_stuff_documents_chain(
+        prompt=prompt,
+        llm=llm,
+        document_prompt=document_prompt,
+    )
+    return create_retrieval_chain(retriever, combine_doc_chain)
+
+
+
+

requirements.txt

@@ -0,0 +1,13 @@
+gradio>=4.38.0
+fastapi>=0.111.0
+uvicorn>=0.28.0
+langchain
+langchain-community
+langchain-qdrant
+huggingface_hub
+qdrant-client
+sentence-transformers
+transformers
+torch
+langchain_groq
+langchain_huggingface