Building AI-Powered Applications with LangChain

LangChain enables developers to build context-aware, reasoning-capable AI agents and chat applications by combining powerful components. In this module, we’ll architect and implement a LangChain-powered Chat API that responds intelligently using memory, tools, and document-based knowledge.

🧠 Application Goals

• Maintain conversation memory (short-term + long-term)
• Use tools like search or calculator
• Retrieve knowledge from documents using Vector DB
• Parse output into clean formats using Output Parsers
• Demonstrate modular structure with reusability

🔧 Architecture Overview

This flowchart illustrates how user inputs flow through a LangChain-based AI system:

• The Agent chooses which tools or chains to call.
• The Memory stores conversation context.
• Documents are queried via Vector Stores.
• Prompt Templates structure the LLM input/output.

📁 File Structure Breakdown

main.py

Handles the FastAPI endpoint and routes user prompts into the LangChain agent pipeline.

from fastapi import FastAPI, Request
from agent import agent_executor

app = FastAPI()

@app.post("/chat")
async def chat(request: Request):
    data = await request.json()
    prompt = data.get("prompt")
    response = agent_executor.run(prompt)
    return {"response": response}

agent.py

Sets up the LangChain agent with tools, memory, and chains.

from langchain.agents import initialize_agent
from tools import tools
from memory import memory
from prompts import SYSTEM_PROMPT
from chains import qa_chain

agent_executor = initialize_agent(
    tools=tools,
    memory=memory,
    agent="chat-conversational-react-description",
    verbose=True,
    handle_parsing_errors=True,
    agent_kwargs={"system_message": SYSTEM_PROMPT}
)

tools.py

Defines external tools the agent can use.

from langchain.tools import Tool
from utils import search_api, calculate

tools = [
    Tool(name="Search", func=search_api, description="For web queries."),
    Tool(name="Calculator", func=calculate, description="Simple math.")
]

memory.py

Stores short-term memory using ConversationBufferMemory.

from langchain.memory import ConversationBufferMemory

memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

prompts.py

Contains templates guiding the LLM behavior.

SYSTEM_PROMPT = "You are a helpful assistant. Always think step-by-step."

chains.py

Encapsulates complex logic like Retrieval QA with a Vector Store.

from langchain.chains import RetrievalQA
from vectordb import retriever

qa_chain = RetrievalQA.from_chain_type(retriever=retriever, chain_type="stuff")

vectordb.py

Loads documents and initializes the vector retriever.

from langchain.vectorstores import FAISS
from langchain.embeddings import OpenAIEmbeddings
from langchain.document_loaders import TextLoader

loader = TextLoader("docs/sample.txt")
docs = loader.load()

db = FAISS.from_documents(docs, OpenAIEmbeddings())
retriever = db.as_retriever()

🧪 Try This Locally

1. Clone repo Github superml.dev Repo
2. Add your OpenAI API key
3. Start with uvicorn main:app
4. Send POST request to /chat with JSON body { "prompt": "Who is Alan Turing?" }

🚀 What You’ve Built

✅ Conversational AI agent
✅ Integrated tools and vector DB
✅ Memory-backed context
✅ Clean structured codebase

Next: Add long-term memory (Redis), advanced chaining (Multi-Input), or UI frontend!