Create main.py
Browse files
main.py
ADDED
@@ -0,0 +1,310 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException, Body, Query, File, UploadFile, Form
|
2 |
+
from fastapi.middleware.cors import CORSMiddleware
|
3 |
+
from pydantic import BaseModel
|
4 |
+
from typing import List, Optional, Dict, Any, Union
|
5 |
+
import uuid
|
6 |
+
import os
|
7 |
+
from dotenv import load_dotenv
|
8 |
+
|
9 |
+
# Load environment variables
|
10 |
+
load_dotenv()
|
11 |
+
|
12 |
+
# Import necessary libraries
|
13 |
+
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
|
14 |
+
from langchain.vectorstores import FAISS
|
15 |
+
from langchain.chains import ConversationalRetrievalChain
|
16 |
+
from langchain_core.prompts import PromptTemplate, ChatPromptTemplate
|
17 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
18 |
+
from langchain_core.documents import Document
|
19 |
+
from langchain_groq import ChatGroq
|
20 |
+
from google import genai
|
21 |
+
from google.genai import types
|
22 |
+
|
23 |
+
# Initialize FastAPI app
|
24 |
+
app = FastAPI(title="RAG System API", description="An API for question answering based on YouTube video content or uploaded video files")
|
25 |
+
|
26 |
+
# Configure CORS
|
27 |
+
app.add_middleware(
|
28 |
+
CORSMiddleware,
|
29 |
+
allow_origins=["*"],
|
30 |
+
allow_credentials=True,
|
31 |
+
allow_methods=["*"],
|
32 |
+
allow_headers=["*"],
|
33 |
+
)
|
34 |
+
|
35 |
+
# Define models
|
36 |
+
class TranscriptionRequest(BaseModel):
|
37 |
+
youtube_url: str
|
38 |
+
|
39 |
+
class QueryRequest(BaseModel):
|
40 |
+
query: str
|
41 |
+
session_id: Optional[str] = None
|
42 |
+
|
43 |
+
class QueryResponse(BaseModel):
|
44 |
+
answer: str
|
45 |
+
session_id: str
|
46 |
+
source_documents: Optional[List[str]] = None
|
47 |
+
|
48 |
+
# Global variables
|
49 |
+
sessions = {}
|
50 |
+
|
51 |
+
# Initialize Google API client
|
52 |
+
def init_google_client():
|
53 |
+
api_key = os.getenv("GOOGLE_API_KEY", "")
|
54 |
+
if not api_key:
|
55 |
+
raise ValueError("GOOGLE_API_KEY environment variable not set")
|
56 |
+
return genai.Client(api_key=api_key)
|
57 |
+
|
58 |
+
# Get LLM
|
59 |
+
def get_llm():
|
60 |
+
"""
|
61 |
+
Returns the language model instance (LLM) using ChatGroq API.
|
62 |
+
The LLM used is Llama 3.1 with a versatile 70 billion parameters model.
|
63 |
+
"""
|
64 |
+
api_key = os.getenv("GROQ_API_KEY", "")
|
65 |
+
if not api_key:
|
66 |
+
raise ValueError("GROQ_API_KEY environment variable not set")
|
67 |
+
|
68 |
+
llm = ChatGroq(
|
69 |
+
model="llama-3.3-70b-versatile",
|
70 |
+
temperature=0,
|
71 |
+
max_tokens=1024,
|
72 |
+
api_key=api_key
|
73 |
+
)
|
74 |
+
return llm
|
75 |
+
|
76 |
+
# Get embeddings
|
77 |
+
def get_embeddings():
|
78 |
+
model_name = "BAAI/bge-small-en"
|
79 |
+
model_kwargs = {"device": "cpu"}
|
80 |
+
encode_kwargs = {"normalize_embeddings": True}
|
81 |
+
embeddings = HuggingFaceBgeEmbeddings(
|
82 |
+
model_name=model_name, model_kwargs=model_kwargs, encode_kwargs=encode_kwargs
|
83 |
+
)
|
84 |
+
return embeddings
|
85 |
+
|
86 |
+
# Create prompt template
|
87 |
+
quiz_solving_prompt = '''
|
88 |
+
You are an assistant specialized in solving quizzes. Your goal is to provide accurate, concise, and contextually relevant answers.
|
89 |
+
Use the following retrieved context to answer the user's question.
|
90 |
+
If the context lacks sufficient information, respond with "I don't know." Do not make up answers or provide unverified information.
|
91 |
+
|
92 |
+
Guidelines:
|
93 |
+
1. Extract key information from the context to form a coherent response.
|
94 |
+
2. Maintain a clear and professional tone.
|
95 |
+
3. If the question requires clarification, specify it politely.
|
96 |
+
|
97 |
+
Retrieved context:
|
98 |
+
{context}
|
99 |
+
|
100 |
+
User's question:
|
101 |
+
{question}
|
102 |
+
|
103 |
+
Your response:
|
104 |
+
'''
|
105 |
+
|
106 |
+
# Create a prompt template to pass the context and user input to the chain
|
107 |
+
user_prompt = ChatPromptTemplate.from_messages(
|
108 |
+
[
|
109 |
+
("system", quiz_solving_prompt),
|
110 |
+
("human", "{question}"),
|
111 |
+
]
|
112 |
+
)
|
113 |
+
|
114 |
+
# Create a chain
|
115 |
+
def create_chain(retriever):
|
116 |
+
llm = get_llm()
|
117 |
+
chain = ConversationalRetrievalChain.from_llm(
|
118 |
+
llm=llm,
|
119 |
+
retriever=retriever,
|
120 |
+
return_source_documents=True,
|
121 |
+
chain_type='stuff',
|
122 |
+
combine_docs_chain_kwargs={"prompt": user_prompt},
|
123 |
+
verbose=False,
|
124 |
+
)
|
125 |
+
return chain
|
126 |
+
|
127 |
+
# Process transcription and prepare RAG system
|
128 |
+
def process_transcription(transcription):
|
129 |
+
# Process the transcription
|
130 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=20)
|
131 |
+
all_splits = text_splitter.split_text(transcription)
|
132 |
+
|
133 |
+
# Create vector store
|
134 |
+
embeddings = get_embeddings()
|
135 |
+
vectorstore = FAISS.from_texts(all_splits, embeddings)
|
136 |
+
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
|
137 |
+
|
138 |
+
# Create a session ID
|
139 |
+
session_id = str(uuid.uuid4())
|
140 |
+
|
141 |
+
# Store session data
|
142 |
+
sessions[session_id] = {
|
143 |
+
"retriever": retriever,
|
144 |
+
"chat_history": [],
|
145 |
+
"transcription": transcription
|
146 |
+
}
|
147 |
+
|
148 |
+
return session_id
|
149 |
+
|
150 |
+
@app.post("/transcribe", response_model=Dict[str, str])
|
151 |
+
async def transcribe_video(request: TranscriptionRequest):
|
152 |
+
"""
|
153 |
+
Transcribe a YouTube video and prepare the RAG system
|
154 |
+
"""
|
155 |
+
try:
|
156 |
+
# Initialize Google API client
|
157 |
+
client = init_google_client()
|
158 |
+
|
159 |
+
# Transcribe the video
|
160 |
+
response = client.models.generate_content(
|
161 |
+
model='models/gemini-2.0-flash',
|
162 |
+
contents=types.Content(
|
163 |
+
parts=[
|
164 |
+
types.Part(text='Transcribe the Video. Write all the things described in the video'),
|
165 |
+
types.Part(
|
166 |
+
file_data=types.FileData(file_uri=request.youtube_url)
|
167 |
+
)
|
168 |
+
]
|
169 |
+
)
|
170 |
+
)
|
171 |
+
|
172 |
+
# Get transcription text
|
173 |
+
transcription = response.candidates[0].content.parts[0].text
|
174 |
+
|
175 |
+
# Process transcription and get session ID
|
176 |
+
session_id = process_transcription(transcription)
|
177 |
+
|
178 |
+
return {"session_id": session_id, "message": "YouTube video transcribed and RAG system prepared"}
|
179 |
+
|
180 |
+
except Exception as e:
|
181 |
+
raise HTTPException(status_code=500, detail=f"Error transcribing video: {str(e)}")
|
182 |
+
|
183 |
+
@app.post("/upload", response_model=Dict[str, str])
|
184 |
+
async def upload_video(file: UploadFile = File(...), prompt: str = Form("Transcribe the Video. Write all the things described in the video")):
|
185 |
+
"""
|
186 |
+
Upload a video file (max 20MB), transcribe it and prepare the RAG system
|
187 |
+
"""
|
188 |
+
try:
|
189 |
+
# Check file size (20MB limit)
|
190 |
+
contents = await file.read()
|
191 |
+
if len(contents) > 20 * 1024 * 1024: # 20MB in bytes
|
192 |
+
raise HTTPException(status_code=400, detail="File size exceeds 20MB limit")
|
193 |
+
|
194 |
+
# Check file type
|
195 |
+
if not file.content_type.startswith('video/'):
|
196 |
+
raise HTTPException(status_code=400, detail="File must be a video")
|
197 |
+
|
198 |
+
# Initialize Google API client
|
199 |
+
client = init_google_client()
|
200 |
+
|
201 |
+
# Transcribe the video
|
202 |
+
response = client.models.generate_content(
|
203 |
+
model='models/gemini-2.0-flash',
|
204 |
+
contents=types.Content(
|
205 |
+
parts=[
|
206 |
+
types.Part(text=prompt),
|
207 |
+
types.Part(
|
208 |
+
inline_data=types.Blob(data=contents, mime_type=file.content_type)
|
209 |
+
)
|
210 |
+
]
|
211 |
+
)
|
212 |
+
)
|
213 |
+
|
214 |
+
# Get transcription text
|
215 |
+
transcription = response.candidates[0].content.parts[0].text
|
216 |
+
|
217 |
+
# Process transcription and get session ID
|
218 |
+
session_id = process_transcription(transcription)
|
219 |
+
|
220 |
+
return {"session_id": session_id, "message": "Uploaded video transcribed and RAG system prepared"}
|
221 |
+
|
222 |
+
except Exception as e:
|
223 |
+
raise HTTPException(status_code=500, detail=f"Error processing uploaded video: {str(e)}")
|
224 |
+
finally:
|
225 |
+
# Reset file pointer
|
226 |
+
await file.seek(0)
|
227 |
+
|
228 |
+
@app.post("/query", response_model=QueryResponse)
|
229 |
+
async def query_system(request: QueryRequest):
|
230 |
+
"""
|
231 |
+
Query the RAG system with a question
|
232 |
+
"""
|
233 |
+
try:
|
234 |
+
session_id = request.session_id
|
235 |
+
|
236 |
+
# Create a new session if none provided
|
237 |
+
if not session_id or session_id not in sessions:
|
238 |
+
raise HTTPException(status_code=404, detail="Session not found. Please transcribe a video first.")
|
239 |
+
|
240 |
+
# Get session data
|
241 |
+
session = sessions[session_id]
|
242 |
+
retriever = session["retriever"]
|
243 |
+
chat_history = session["chat_history"]
|
244 |
+
|
245 |
+
# Create chain
|
246 |
+
chain = create_chain(retriever)
|
247 |
+
|
248 |
+
# Query the chain
|
249 |
+
result = chain({"question": request.query, "chat_history": chat_history})
|
250 |
+
|
251 |
+
# Update chat history
|
252 |
+
chat_history.append((request.query, result["answer"]))
|
253 |
+
|
254 |
+
# Prepare source documents
|
255 |
+
source_docs = [doc.page_content[:100] + "..." for doc in result.get("source_documents", [])]
|
256 |
+
|
257 |
+
return {
|
258 |
+
"answer": result["answer"],
|
259 |
+
"session_id": session_id,
|
260 |
+
"source_documents": source_docs
|
261 |
+
}
|
262 |
+
|
263 |
+
except Exception as e:
|
264 |
+
raise HTTPException(status_code=500, detail=f"Error querying system: {str(e)}")
|
265 |
+
|
266 |
+
@app.get("/sessions/{session_id}", response_model=Dict[str, Any])
|
267 |
+
async def get_session_info(session_id: str):
|
268 |
+
"""
|
269 |
+
Get information about a specific session
|
270 |
+
"""
|
271 |
+
if session_id not in sessions:
|
272 |
+
raise HTTPException(status_code=404, detail="Session not found")
|
273 |
+
|
274 |
+
session = sessions[session_id]
|
275 |
+
|
276 |
+
return {
|
277 |
+
"session_id": session_id,
|
278 |
+
"chat_history_length": len(session["chat_history"]),
|
279 |
+
"transcription_preview": session["transcription"][:200] + "..."
|
280 |
+
}
|
281 |
+
|
282 |
+
@app.delete("/sessions/{session_id}")
|
283 |
+
async def delete_session(session_id: str):
|
284 |
+
"""
|
285 |
+
Delete a session
|
286 |
+
"""
|
287 |
+
if session_id not in sessions:
|
288 |
+
raise HTTPException(status_code=404, detail="Session not found")
|
289 |
+
|
290 |
+
del sessions[session_id]
|
291 |
+
return {"message": f"Session {session_id} deleted successfully"}
|
292 |
+
|
293 |
+
@app.get("/")
|
294 |
+
async def root():
|
295 |
+
"""
|
296 |
+
API root endpoint
|
297 |
+
"""
|
298 |
+
return {
|
299 |
+
"message": "Video Transcription and QA API",
|
300 |
+
"endpoints": {
|
301 |
+
"/transcribe": "Transcribe YouTube videos",
|
302 |
+
"/upload": "Upload and transcribe video files (max 20MB)",
|
303 |
+
"/query": "Query the RAG system",
|
304 |
+
"/sessions/{session_id}": "Get session information",
|
305 |
+
}
|
306 |
+
}
|
307 |
+
|
308 |
+
if __name__ == "__main__":
|
309 |
+
import uvicorn
|
310 |
+
uvicorn.run(app, host="0.0.0.0", port=8000)
|