48 lines
1.4 KiB
Python
48 lines
1.4 KiB
Python
import os
|
|
from fastapi import FastAPI
|
|
from langchain_community.utilities import SearxSearchWrapper
|
|
from langchain_community.vectorstores import Qdrant
|
|
from langchain_community.embeddings import HuggingFaceEmbeddings
|
|
from langchain_community.llms import LLaMAServer
|
|
from langchain.schema import Document
|
|
|
|
# Environment variables
|
|
SEARXNG_URL = os.environ["SEARXNG_URL"]
|
|
QDRANT_URL = os.environ["QDRANT_URL"]
|
|
LLAMA_SERVER_URL = os.environ["LLAMA_SERVER_URL"]
|
|
|
|
# Initialize FastAPI
|
|
app = FastAPI()
|
|
|
|
# Initialize components
|
|
searx = SearxSearchWrapper(searx_host=SEARXNG_URL)
|
|
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
|
qdrant = Qdrant(
|
|
url=QDRANT_URL,
|
|
collection_name="search_vectors",
|
|
embeddings=embedding_model
|
|
)
|
|
llm = LLaMAServer(endpoint_url=LLAMA_SERVER_URL)
|
|
|
|
@app.post("/process")
|
|
async def process(query: str):
|
|
# Step 1: Search with SearXNG
|
|
raw_results = searx.run(query)
|
|
|
|
# Step 2: Convert to LangChain Documents
|
|
docs = [Document(page_content=result) for result in raw_results.split("\n") if result.strip()]
|
|
|
|
# Step 3: Embed and store in Qdrant
|
|
qdrant.add_documents(docs)
|
|
|
|
# Step 4: Query Qdrant for similar vectors
|
|
matches = qdrant.similarity_search(query, k=3)
|
|
|
|
# Step 5: Combine raw + vector results
|
|
combined = "\n".join([doc.page_content for doc in docs + matches])
|
|
|
|
# Step 6: Send to LLaMA-Server
|
|
response = llm.invoke(combined)
|
|
|
|
return {"response": response}
|