Prepare project for Hugging Face deployment

.gitignore

@@ -12,6 +12,9 @@ dist
 dist-ssr
 *.local
 .env
+back/data/uploads/
+back/chroma_db/
+back/chats.db
 
 # Editor directories and files
 .vscode/*

back/.gitignore

@@ -16,6 +16,8 @@ vector_db.pkl
 # Ma'lumotlar bazalari va Vektor do'konlari
 chroma_db/
 chats.db
+data/
+data/uploads
 
 # Caches
 .cache/

back/requirements.txt

@@ -10,4 +10,6 @@ requests
 google-genai
 python-multipart
 gradio
-
+transformers>=4.37.0
+sentence-transformers
+openai

back/server.py

@@ -1,6 +1,5 @@
 import os
-import json
-import pickle
+from openai import OpenAI
 import numpy as np
 import sqlite3
 import uuid
@@ -8,26 +7,25 @@ from datetime import datetime
 from typing import List, Optional
 from dotenv import load_dotenv
 from pypdf import PdfReader
-from google import genai
-from google.genai import types
-from sklearn.metrics.pairwise import cosine_similarity
-# import chromadb - Moved to try/except block below
-
+from pathlib import Path
+import re
 
 # ==============================
 # 0. Sozlamalar
 # ==============================
-load_dotenv()
 
-API_KEY = os.getenv("GEMINI_API_KEY")
-if not API_KEY:
-    raise RuntimeError("GEMINI_API_KEY topilmadi!")
 
-client = genai.Client(api_key=API_KEY)
-# Chat modeli (tez va arzon variant)
-GEMINI_CHAT_MODEL = "gemini-1.5-flash"
-# Embedding modeli
-GEMINI_EMBED_MODEL = "text-embedding-004"
+env_path = Path(__file__).resolve().parent / ".env"
+load_dotenv(env_path, override=True)
+
+OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+if not OPENAI_API_KEY:
+    raise RuntimeError("OPENAI_API_KEY topilmadi")
+
+openai_client = OpenAI(api_key=OPENAI_API_KEY)
+
+OPENAI_EMBED_MODEL = "text-embedding-3-small"
+OPENAI_CHAT_MODEL = "gpt-4o-mini"
 
 CHROMA_DIR = "./chroma_db"
 CHAT_DB_PATH = "./chats.db"
@@ -46,23 +44,79 @@ except Exception as e:
     collection = None
     RAG_AVAILABLE = False
 
+print("RAG_AVAILABLE:", RAG_AVAILABLE)
+print("COLLECTION COUNT:", collection.count() if collection else 0)
 
 # ==============================
 # 1. PDF -> TEXT
 # ==============================
+# def load_pdf(path: str) -> str:
+#     reader = PdfReader(path)
+#     text = ""
+#     for page in reader.pages:
+#         page_text = page.extract_text()
+#         if page_text:
+#             text += page_text + "\n"
+#     return text
+
+def fix_spaced_text(line: str) -> str:
+    if re.fullmatch(r'(?:[A-Za-z]\s+){3,}[A-Za-z]?', line.strip()):
+        return line.replace(" ", "")
+    return line
+
 def load_pdf(path: str) -> str:
     reader = PdfReader(path)
-    text = ""
+    lines = []
+
     for page in reader.pages:
         page_text = page.extract_text()
-        if page_text:
-            text += page_text + "\n"
-    return text
+        if not page_text:
+            continue
+
+        for line in page_text.splitlines():
+            cleaned = fix_spaced_text(line)
+            if cleaned:
+                lines.append(cleaned)
+
+    text = "\n".join(lines)
+
+    text = re.sub(r'(?<=\w)\s*@\s*(?=\w)', '@', text)
+    text = re.sub(r'(?<=\w)\s*\.\s*(?=\w)', '.', text)
+    text = re.sub(r'\n{3,}', '\n\n', text)
+    text = re.sub(r'[ \t]{2,}', ' ', text)
+
+    return text.strip()
+
+def extract_email_from_context(context_list: List[str]) -> Optional[str]:
+    text = "\n".join(context_list)
+
+    text = re.sub(r'(?<=\w)\s*@\s*(?=\w)', '@', text)
+    text = re.sub(r'(?<=\w)\s*\.\s*(?=\w)', '.', text)
+    text = re.sub(r'(?<=\w)\s+(?=\w@)', '', text)
+    text = re.sub(r'(?<=@)\s+(?=\w)', '', text)
+
+    match = re.search(r'[\w\.-]+@[\w\.-]+\.\w+', text)
+    return match.group(0) if match else None
+
+
+def extract_project_lines(context_list: List[str]) -> List[str]:
+    text = "\n".join(context_list)
+    lines = [line.strip() for line in text.splitlines() if line.strip()]
+
+    keywords = ["project", "cs core", "trusty", "mily", "corporate solutions"]
+    result = []
+
+    for line in lines:
+        low = line.lower()
+        if any(k in low for k in keywords):
+            result.append(line)
+
+    return result[:8]
 
 # ==============================
 # 2. TEXT -> CHUNKS
 # ==============================
-def chunk_text(text, chunk_size=300, overlap=200):
+def chunk_text(text, chunk_size=500, overlap=100):
     chunks = []
     start = 0
     while start < len(text):
@@ -74,16 +128,34 @@ def chunk_text(text, chunk_size=300, overlap=200):
 # ==============================
 # 3. CHUNKS -> EMBEDDINGS
 # ==============================
-def embed_texts(texts):
-    # Ollama embedding for a list of texts
-    embeddings = []
-    for text in texts:
-        response = client.models.embed_content(
-            model=GEMINI_EMBED_MODEL,
-            contents=text
-        )
-        embeddings.append(response.embedding.values)
-    return embeddings
+# def embed_texts(texts):
+#     # Ollama embedding for a list of texts
+#     embeddings = []
+#     for text in texts:
+#         response = client.models.embed_content(
+#             model=GEMINI_EMBED_MODEL,
+#             contents=text
+#         )
+#         embeddings.append(response.embeddings[0].values)
+#     return embeddings
+
+# qwen
+# def embed_texts(texts):
+#     embeddings = embedder.encode(texts, convert_to_numpy=True, normalize_embeddings=True)
+#     return embeddings.tolist()
+
+OPENAI_EMBED_MODEL = "text-embedding-3-small"
+
+def embed_texts(texts: List[str]) -> List[List[float]]:
+    cleaned = [t.strip() for t in texts if t and t.strip()]
+    if not cleaned:
+        return []
+
+    response = openai_client.embeddings.create(
+        model=OPENAI_EMBED_MODEL,
+        input=cleaned
+    )
+    return [item.embedding for item in response.data]
 
 
 # ==============================
@@ -126,7 +198,7 @@ def delete_from_chroma(doc_id):
 # ==============================
 # Helper for RAG Tool
 # ==============================
-def find_context(query, top_k=3):
+def find_context(query, top_k=4):
     if not RAG_AVAILABLE: return []
     try:
         query_embedding = embed_texts([query])[0]
@@ -156,9 +228,6 @@ def retrieve_documents(query: str) -> str:
 # ... (init_db, CRUD, etc - skipped for brevity in tool call logic, assuming target content matches)
 
 
-
-
-
 # ==============================
 # 6. CHAT & DOCUMENT DATABASE SETUP
 # ==============================
@@ -366,7 +435,7 @@ def get_chat_messages(chat_id: str) -> List[dict]:
 # 9. RAG-AWARE GENERATION
 # ==============================
 
-from tools import calculate_expression, get_current_weather
+# from tools import calculate_expression, get_current_weather
 # from google.genai.types import Tool, GenerateContentConfig, FunctionDeclaration
 
 
@@ -374,265 +443,88 @@ from tools import calculate_expression, get_current_weather
 # 9. RAG-AWARE GENERATION
 # ==============================
 
-SYSTEM_PROMPT = """You are a helpful assistant. 
-- Answer general greetings (like 'hi', 'hello') directly and briefly.
-- Use `retrieve_documents` ONLY for questions about uploaded files.
-- Use `calculate_expression` ONLY for math.
-- Use `get_current_weather` ONLY for weather questions.
-DO NOT use tools for simple conversation.
-"""
+SYSTEM_PROMPT = (
+    "You are a document-grounded assistant. "
+    "Answer the user's question using only the provided document context and relevant chat history. "
+    "Do not guess, do not invent facts, and do not add information that is not supported by the document context. "
+    "If the answer is not clearly available in the provided context, say exactly: "
+    "'The exact answer is not clearly available in the document.' "
+    "When the document contains the answer, provide a complete and accurate response with all relevant details found in the context. "
+    "Preserve important names, numbers, dates, email addresses, links, titles, and technical terms exactly as they appear in the document whenever possible."
+)
 
+# SYSTEM_PROMPT = """You are a helpful assistant. 
+# - Answer general greetings (like 'hi', 'hello') directly and briefly.
+# - Use `retrieve_documents` ONLY for questions about uploaded files.
+# - Use `calculate_expression` ONLY for math.
+# - Use `get_current_weather` ONLY for weather questions.
+# DO NOT use tools for simple conversation.
+# """
 
+OPENAI_CHAT_MODEL = "gpt-4o-mini"
 
 def generate_rag_response(question: str, context_list: List[str], chat_history: List[dict]) -> str:
-    """
-    Generate a response using RAG with conversation history and tools.
-    """
-    # --- ROUTING LAYER (For 1B model stability) ---
-    is_greeting = question.lower().strip() in ["hi", "hello", "hey", "salom", "qalay", "howdy"]
-    is_very_short = len(question.strip()) < 10
-    
-    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
-    history_messages = chat_history[-10:] if len(chat_history) > 10 else chat_history
-    for msg in history_messages:
-        messages.append({"role": msg["role"], "content": msg["content"]})
-    messages.append({"role": "user", "content": question})
-
-    # If it's just a greeting, don't even show tools to the 1B model
-    if is_greeting or (is_very_short and not any(char.isdigit() for char in question)):
-        print(f"--- Routing: Simple greeting detected. Skipping tools. ---")
-        try:
-            response = client.models.generate_content(
-                model=GEMINI_CHAT_MODEL,
-                contents=question,
-                config=types.GenerateContentConfig(
-                    system_instruction="You are a friendly assistant. Greet the user normally and briefly.",
-                    temperature=0,
-    ),
-                # messages=[{"role": "system", "content": "You are a friendly assistant. Greet the user normally and briefly."}, {"role": "user", "content": question}],
-                # options={'temperature': 0}
-            )
-            # return response.text
-            return (response.text or "Hello! How can I help you todayyy?").strip()
-        except Exception as e:
-            print(f"Gemini Routing Error: {e}")
-            return "Hello! How can I help you today?"
-
-    # --- STANDARD TOOL CALLING LAYER ---
-    # tools =[
-    #     {
-    #         'type': 'function',
-    #         'function': {
-    #             'name': 'calculate_expression',
-    #             'description': 'Solve arithmetic math problems (e.g. 2+2).',
-    #             'parameters': {
-    #                 'type': 'object',
-    #                 'properties': {
-    #                     'expression': {'type': 'string', 'description': 'The math expression'},
-    #                 },
-    #                 'required': ['expression'],
-    #             },
-    #         },
-    #     },
-    #     {
-    #         'type': 'function',
-    #         'function': {
-    #             'name': 'get_current_weather',
-    #             'description': 'Get the current weather for a city.',
-    #             'parameters': {
-    #                 'type': 'object',
-    #                 'properties': {
-    #                     'location': {'type': 'string', 'description': 'City name'},
-    #                 },
-    #                 'required': ['location'],
-    #             },
-    #         },
-    #     },
-    #     {
-    #         'type': 'function',
-    #         'function': {
-    #             'name': 'retrieve_documents',
-    #             'description': 'Search for information in uploaded PDF documents.',
-    #             'parameters': {
-    #                 'type': 'object',
-    #                 'properties': {
-    #                     'query': {'type': 'string', 'description': 'The search query'},
-    #                 },
-    #                 'required': ['query'],
-    #             },
-    #         },
-    #     },
-    # ]
-
-    tool = types.Tool(function_declarations=[
-    types.FunctionDeclaration(
-        name="calculate_expression",
-        description="Solve arithmetic math problems (e.g. 2+2).",
-        parameters_json_schema={
-            "type": "object",
-            "properties": {
-                "expression": {"type": "string", "description": "The math expression"},
-            },
-            "required": ["expression"],
-        },
-    ),
-    types.FunctionDeclaration(
-        name="get_current_weather",
-        description="Get the current weather for a city.",
-        parameters_json_schema={
-            "type": "object",
-            "properties": {
-                "location": {"type": "string", "description": "City name"},
-            },
-            "required": ["location"],
-        },
-    ),
-    types.FunctionDeclaration(
-        name="retrieve_documents",
-        description="Search for information in uploaded PDF documents.",
-        parameters_json_schema={
-            "type": "object",
-            "properties": {
-                "query": {"type": "string", "description": "The search query"},
-            },
-            "required": ["query"],
-        },
-    ),
-])
+    # context_text = "\n\n---\n\n".join(context_list[:2]) if context_list else "No relevant context found."
+    context_text = "\n\n---\n\n".join(context_list) if context_list else "No relevant context found."
 
+    history_text = ""
+    for msg in (chat_history[-3:] if len(chat_history) > 3 else chat_history):
+        history_text += f"{msg['role'].upper()}: {msg['content']}\n"
 
-    try:
-        print(f"--- Sending to Ollama: {question} ---")
-        available_functions = {
-        "calculate_expression": calculate_expression,
-        "get_current_weather": get_current_weather,
-        "retrieve_documents": retrieve_documents,
-}
-
-# Gemini uchun promptni “system + history + user” ko‘rinishida bitta textga yig’amiz
-        system_text = SYSTEM_PROMPT
-        history_text = ""
-        for msg in (chat_history[-10:] if len(chat_history) > 10 else chat_history):
-            history_text += f"{msg['role'].upper()}: {msg['content']}\n"
-
-        user_text = f"{history_text}\nUSER: {question}".strip()
-        response = client.models.generate_content(
-            model=GEMINI_CHAT_MODEL,
-            contents=user_text,
-            config=types.GenerateContentConfig(
-            system_instruction=system_text,
-            tools=[tool],
-            temperature=0,
-            automatic_function_calling=types.AutomaticFunctionCallingConfig(disable=False),
-    ),
-            # messages=messages,
-            # tools=tools,
-            # options={'temperature': 0}
-        )
+    prompt = f"""Document context:
+{context_text}
 
-        print(f"Raw Ollama response content: {response.text}")
+Chat history:
+{history_text}
 
-        # 1. Native Tool Calls
-        if response.function_calls:
-            print(f"Native tool calls detected: {response.function_calls}")
-            available_functions = {
-                'calculate_expression': calculate_expression,
-                'get_current_weather': get_current_weather,
-                'retrieve_documents': retrieve_documents,
-            }
-            
-            # messages.append(response.message)
-
-            tool_outputs = []
-
-            for call in response.function_calls:
-                # google-genai SDK da odatda shu ko‘rinish:
-                func_name = getattr(call, "name", None)
-                func_args = getattr(call, "args", None)
-                    
-                # fallback (agar boshqa format bo‘lsa):
-                if func_name is None and hasattr(call, "function"):
-                    func_name = getattr(call.function, "name", None)
-                    func_args = getattr(call.function, "arguments", None)
-
-                if isinstance(func_args, str):
-                    try:
-                        func_args = json.loads(func_args)
-                    except Exception:
-                        func_args = {}
-
-                if func_args is None:
-                    func_args = {}
-
-                if func_name not in available_functions:
-                    tool_outputs.append(f"{func_name}: Unknown tool")
-                    continue
-                
-                try:
-                    result = available_functions[func_name](**func_args)
-                except Exception as e:
-                    result = f"Tool error: {e}"
-
-                tool_outputs.append(f"{func_name}({func_args}) => {result}")
-
-            followup_prompt = (
-                f"{user_text}\n\n"
-                "Tool results:\n"
-                + "\n".join(tool_outputs)
-                + "\n\nNow answer the user using the tool results."
-    )
+User question:
+{question}
+"""
 
-            final_response = client.models.generate_content(
-                model=GEMINI_CHAT_MODEL,
-                contents=followup_prompt,
-                config=types.GenerateContentConfig(
-                    system_instruction=system_text,
-                    temperature=0,
-        ),
+    response = openai_client.responses.create(
+        model=OPENAI_CHAT_MODEL,
+        input=[
+            {
+                "role": "system",
+                "content": [
+                    {
+                        "type": "input_text",
+                        "text": (
+                            "You are a document-grounded assistant. "
+                            "Answer ONLY from the provided document context. "
+                            "Do not guess. Do not rewrite names, emails, project names, companies, locations, or technologies. "
+                            "If the exact answer is not clearly available in the document, say exactly: "
+                            "'The exact answer is not clearly available in the document.' "
+                            "Keep the answer short and factual."
+                        ),
+                        # "text": (
+                        #     "You are a document-grounded assistant. "
+                        #     "Answer ONLY from the provided document context. "
+                        #     "Do not guess. Do not rewrite names, emails, project names, companies, locations, or technologies. "
+                        #     "If the exact answer is not clearly available in the document, say exactly: "
+                        #     "'The exact answer is not clearly available in the document.' "
+                        #     "Keep the answer short and factual."
+                        # ),
+                    }
+                ],
+            },
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "input_text",
+                        "text": prompt,
+                    }
+                ],
+            },
+        ],
     )
-            return (final_response.text or "").strip()
-        
-        # 2. Hard Fallback for math/tool hallucinations
-        content = response.text.strip() if response.text else ""
-        hallucination_keywords = ["calculate_expression", "syntax error", "expression", "parameters"]
-        
-        if any(kw in content.lower() for kw in hallucination_keywords) and not any(char.isdigit() for char in question):
-             print(f"Detected tool hallucination in text: {content[:50]}...")
-             retry_response = client.models.generate_content(
-                model=GEMINI_CHAT_MODEL,
-                messages=[{"role": "system", "content": "You are a helpful assistant. Provide a natural response without mentioning tools or syntax."}, {"role": "user", "content": question}],
-                options={'temperature': 0}
-             )
-             return retry_response.text
-
-        return content
-
-    except Exception as e:
-        print(f"Ollama Error: {e}")
-        return f"Error generation response: {str(e)}"
-
-
 
+    return (response.output_text or "").strip() or "The exact answer is not clearly available in the document."
 
 # Legacy function - kept for backwards compatibility
 def ask_gemini(question, context_list):
-    context = "\n\n".join(context_list)
-    prompt = f"""Quyidagi context ma'lumotlaridan foydalanib savolga javob ber. 
-Faqat context ichidagi ma'lumotni ishlat.
-
-Context:
-{context}
-
-Savol: {question}"""
-
-    response = client.models.generate_content(
-        # messages=[{'role': 'user', 'content': prompt}]
-        model=GEMINI_CHAT_MODEL,
-        contents=prompt,
-        config=types.GenerateContentConfig(temperature=0),
-    )
-    return (response.text or "").strip()
-
+    return generate_rag_response(question, context_list, [])
 
 # ==============================
 # 10. MAIN PROCESS (PDF Processing)
@@ -650,7 +542,6 @@ from fastapi import FastAPI, HTTPException, UploadFile, File
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 
-
 app = FastAPI(title="RAG Chat API", version="2.0.0")
 app.add_middleware(
     CORSMiddleware,
@@ -676,94 +567,24 @@ class SendMessageRequest(BaseModel):
 class UpdateChatRequest(BaseModel):
     title: str
 
-
-# ==============================
-# 14.5 DOCUMENT ENDPOINTS
-# ==============================
-
-@app.get("/documents")
-async def list_documents():
-    return {"documents": get_all_documents()}
-
-@app.post("/documents")
-async def upload_document(file: UploadFile = File(...)):
-    if not RAG_AVAILABLE:
-        raise HTTPException(status_code=503, detail="RAG system unavailable (Python 3.14 incompatibility).")
-
-    if not file.filename.endswith('.pdf'):
-        raise HTTPException(status_code=400, detail="Only PDF files are allowed")
-    
-    # Save file temporarily
-    os.makedirs("data/uploads", exist_ok=True)
-    file_path = f"data/uploads/{uuid.uuid4()}_{file.filename}"
-    
-    try:
-        with open(file_path, "wb") as f:
-            content = await file.read()
-            f.write(content)
-            
-        # Process PDF
-        text = load_pdf(file_path)
-        if not text.strip():
-            raise HTTPException(status_code=400, detail="Could not extract text from PDF")
-            
-        chunks = chunk_text(text)
-        embeddings = embed_texts(chunks)
-        
-        # Save Metadata
-        doc = create_document_record(file.filename)
-        
-        # Save Vectors
-        save_to_chroma(chunks, embeddings, doc["id"])
-        
-        # Cleanup file (optional, keeping it for now in case needed, or delete)
-        # os.remove(file_path) 
-        
-        return doc
-        
-    except Exception as e:
-        if os.path.exists(file_path):
-            os.remove(file_path)
-        raise HTTPException(status_code=500, detail=f"Processing failed: {str(e)}")
-
-@app.delete("/documents/{doc_id}")
-async def delete_document(doc_id: str):
-    if not RAG_AVAILABLE:
-         # Still allow deleting from DB, just skip vector delete
-         pass
-        
-    success = delete_document_record(doc_id)
-    if not success:
-        raise HTTPException(status_code=404, detail="Document not found")
-    
-    # Remove from Vector DB
-    if RAG_AVAILABLE:
-        delete_from_chroma(doc_id)
-    
-    return {"message": "Document deleted successfully"}
-
-
 # ==============================
 # 13. LEGACY ENDPOINT (backwards compatibility)
 # ==============================
 
 @app.post("/ask")
 async def ask_question(req: QuestionRequest):
-    """Legacy endpoint - still functional for backwards compatibility"""
     question = req.question.strip()
     if not question:
         raise HTTPException(status_code=400, detail="Savol bo'sh bo'lishi mumkin emas")
 
-    # This endpoint now relies on documents already in ChromaDB, not auto-loading data.pdf
     if collection.count() == 0:
         raise HTTPException(status_code=404, detail="No documents loaded into the system. Please upload PDFs first.")
 
-    relevant_chunks = find_context(question, top_k=3)
-    answer = ask_gemini(question, relevant_chunks)
+    relevant_chunks = find_context(question, top_k=4)
+    answer = generate_rag_response(question, relevant_chunks, [])
 
     return {"answer": answer}
 
-
 # ==============================
 # 14. CHAT ENDPOINTS
 # ==============================
@@ -817,61 +638,51 @@ async def remove_chat(chat_id: str):
 async def list_documents():
     return {"documents": get_all_documents()}
 
+
 @app.post("/documents")
 async def upload_document(file: UploadFile = File(...)):
     if not RAG_AVAILABLE:
-        raise HTTPException(status_code=503, detail="RAG system unavailable (Python 3.14 incompatibility).")
+        raise HTTPException(status_code=503, detail="RAG system unavailable.")
 
-    if not file.filename.endswith('.pdf'):
+    if not file.filename.endswith(".pdf"):
         raise HTTPException(status_code=400, detail="Only PDF files are allowed")
-    
-    # Save file temporarily
+
     os.makedirs("data/uploads", exist_ok=True)
     file_path = f"data/uploads/{uuid.uuid4()}_{file.filename}"
-    
+
     try:
         with open(file_path, "wb") as f:
             content = await file.read()
             f.write(content)
-            
-        # Process PDF
+
         text = load_pdf(file_path)
         if not text.strip():
             raise HTTPException(status_code=400, detail="Could not extract text from PDF")
-            
+
         chunks = chunk_text(text)
         embeddings = embed_texts(chunks)
-        
-        # Save Metadata
+
         doc = create_document_record(file.filename)
-        
-        # Save Vectors
         save_to_chroma(chunks, embeddings, doc["id"])
-        
-        # Cleanup file (optional, keeping it for now in case needed, or delete)
-        # os.remove(file_path) 
-        
+
         return doc
-        
+
     except Exception as e:
         if os.path.exists(file_path):
             os.remove(file_path)
         raise HTTPException(status_code=500, detail=f"Processing failed: {str(e)}")
 
+
 @app.delete("/documents/{doc_id}")
 async def delete_document(doc_id: str):
     success = delete_document_record(doc_id)
     if not success:
         raise HTTPException(status_code=404, detail="Document not found")
-    
-    # Remove from Vector DB
+
     if RAG_AVAILABLE:
         delete_from_chroma(doc_id)
-    
-    return {"message": "Document deleted successfully"}
-
-
 
+    return {"message": "Document deleted successfully"}
 # ==============================
 # 15. MESSAGE ENDPOINTS
 # ==============================
@@ -889,54 +700,48 @@ async def get_messages(chat_id: str):
 
 @app.post("/chats/{chat_id}/messages")
 async def send_message(chat_id: str, req: SendMessageRequest):
-    """
-    Send a message and get RAG-powered response.
-    
-    This endpoint:
-    1. Saves the user message
-    2. Retrieves relevant context from vector DB
-    3. Generates response using chat history + RAG context
-    4. Saves and returns the assistant response
-    """
     chat = get_chat_by_id(chat_id)
     if not chat:
         raise HTTPException(status_code=404, detail="Chat not found")
-    
+
     content = req.content.strip()
     if not content:
         raise HTTPException(status_code=400, detail="Message content cannot be empty")
-    
-    # NO AUTO_LOAD of data.pdf anymore. RAG uses whatever is in Chroma.
-    
-    # Save user message
+
     user_message = add_message(chat_id, "user", content)
-    
-    # Update chat title if this is the first message
+
     messages = get_chat_messages(chat_id)
-    if len(messages) == 1:  # Only the user message we just added
-        # Generate title from first message (truncate if too long)
+    if len(messages) == 1:
         title = content[:50] + "..." if len(content) > 50 else content
         update_chat_title(chat_id, title)
-    
-    # Get relevant context from RAG (Legacy/Fallback)
-    # The model will now use retrieve_documents tool if needed
-    relevant_chunks = []
-    
-    # Get chat history (excluding the message we just added for cleaner history)
+
+    relevant_chunks = find_context(content, top_k=4)
+    print("RELEVANT CHUNKS:", relevant_chunks)
+
     chat_history = messages[:-1] if len(messages) > 1 else []
-    
-    # Generate RAG-powered response
-    response_text = generate_rag_response(content, relevant_chunks, chat_history)
-    
-    # Save assistant message
+    lower_content = content.lower()
+
+    if "email" in lower_content or "e-mail" in lower_content or "gmail" in lower_content:
+        email = extract_email_from_context(relevant_chunks)
+        response_text = email if email else "The exact answer is not clearly available in the document."
+
+    elif "project" in lower_content:
+        project_lines = extract_project_lines(relevant_chunks)
+        if project_lines:
+            response_text = "\n".join(project_lines)
+        else:
+            response_text = generate_rag_response(content, relevant_chunks, chat_history)
+
+    else:
+        response_text = generate_rag_response(content, relevant_chunks, chat_history)
+
     assistant_message = add_message(chat_id, "assistant", response_text)
-    
+
     return {
         "user_message": user_message,
         "assistant_message": assistant_message
     }
 
-
 # ==============================
 # 16. SERVER STARTUP
 # ==============================
@@ -977,4 +782,4 @@ if __name__ == "__main__":
             print(f"Failed to auto-load data.pdf: {e}")
 
     import uvicorn
-    uvicorn.run("server:app", host="0.0.0.0", port=4000, reload=True)
+    uvicorn.run("server:app", host="0.0.0.0", port=4000)

services/api.ts

@@ -1,7 +1,7 @@
 // Chat API Service
 import { Chat, Message, Document } from '../types';
 
-const API_BASE = 'http://localhost:4000';
+const API_BASE = 'http://localhost:8000';
 
 // Create a new chat
 export async function createChat(title?: string): Promise<Chat> {

vite.config.ts

@@ -6,7 +6,7 @@ export default defineConfig(({ mode }) => {
     const env = loadEnv(mode, '.', '');
     return {
       server: {
-        port: 3000,
+        port: 4000,
         host: '0.0.0.0',
       },
       plugins: [react()],