DocStrange API + LangChain: Build High-Quality Document RAG in Minutes

Copy

pip install langchain_community
pip install chromadb
pip install langchain_openai
pip install langchain-huggingface langchain-chroma

Copy

import requests


    url = "https://extraction-api.nanonets.com/api/v1/extract/sync"
    
    
    files = { "file": ("example-file", open("example-file", "rb")) }
    payload = {
        "output_format": "markdown",
        "file_url": "",
        "file_base64": "",
        "custom_instructions": "",
        "prompt_mode": "append",
        "json_options": "",
        "csv_options": "",
        "include_metadata": ""
    }
    headers = {"Authorization": "Bearer {API_KEY}"}
    
    
    resp = requests.post(url, data=payload, files=files, headers=headers)
    
    
    resp.raise_for_status()
    
    
    markdown = resp.json()["result"]["markdown"]["content"]
    with open("annual_report.md", "w", encoding="utf-8") as f:
       f.write(markdown)
    
    
    print("✅ Extracted and saved: annual_report.md")

Copy

from pathlib import Path
from langchain_core.documents import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_chroma import Chroma


# Load the Markdown file produced by DocStrange
text = Path("annual_report.md").read_text(encoding="utf-8")
docs = [Document(page_content=text, metadata={"source": "annual_report.pdf"})]

# Split for embeddings
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
chunks = splitter.split_documents(docs)

# Create embeddings
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Build Chroma vector index (persistent)
persist_dir = "indexes/annual_report_chroma"
collection_name = "annual_report_2023"

vectorstore = Chroma.from_documents(
    documents=chunks,
    embedding=embeddings,
    persist_directory=persist_dir,
    collection_name=collection_name,
)

print("✅ Built Chroma index from extracted Markdown")

Copy

# ---------- RAG query pipeline ----------
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI

QUESTION = "Summarize key revenue insights from the 2023 section."

# Reopen the persisted Chroma store
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
vs = Chroma(
    embedding_function=embeddings,
    persist_directory=persist_dir,
    collection_name=collection_name,
)

retriever = vs.as_retriever(search_kwargs={"k": 5})

prompt = ChatPromptTemplate.from_template("""
You are a financial analyst. Use the provided context to answer the question.

Question: {question}

Context:
{context}

Answer:
""")

llm = ChatOpenAI(model="gpt-4.1", api_key="YOUR_OPENAI_API_KEY")

def format_docs(docs):
    return "\n\n---\n\n".join(d.page_content for d in docs)

chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

print(chain.invoke(QUESTION))