🔹 Summary

Set up LangChain in the backend to handle AI-related operations (system instructions, prompts, chains, etc.).
This includes initializing the LLM client, creating a reusable LangChain service, and preparing basic chains that can be used across the project.

👤 User Story

As a developer,
I want a centralized LangChain setup in the backend,
So that I can easily build and reuse AI features (instruction handling, chat, RAG, etc.) without rewriting LLM logic each time.

✅ Scope / Requirements

1. Environment & Config

Add required environment variables:

OPENAI_API_KEY (or other provider key if used)

Create a config file/module to read AI-related env vars.

Make model name configurable, e.g.:

OPENAI_MODEL=gpt-4.1-mini (or similar)

2. LangChain Initialization

Install required dependencies:

@langchain/openai

langchain

Create a dedicated file/module, e.g.:

src/lib/langchainClient.js or src/services/langchain.service.js

Inside it:

Initialize ChatOpenAI instance.

Export a reusable model instance.

Example (for dev reference, not required in issue body):

import { ChatOpenAI } from "@langchain/openai";

export const llm = new ChatOpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  model: process.env.OPENAI_MODEL || "gpt-4.1-mini",
});

3. Basic Prompt + Chain Setup

Create a simple system + user prompt flow using ChatPromptTemplate.

Wrap model + prompt into a RunnableSequence chain.

Expose a function, e.g.:

runSimplePrompt({ system, user })

This should:

Accept parameters (input text, system instructions).

Return message.content as plain text.

4. Integration Point

Prepare this service so it can later be used with:

System instructions from DB (Excel/Word upload feature).

Future endpoints like /api/ai/chat, /api/ai/generate, etc.

Keep the service decoupled from HTTP layer (no req, res inside it).

🔍 Expected Flow

Backend starts.

LangChain client initializes using env vars.

Other modules import llm or runSimplePrompt() from the LangChain service.

When called, the service:

Builds prompt.

Invokes model.

Returns response as text/JSON to the caller.

✅ Acceptance Criteria

@langchain/openai and langchain are installed and listed in package.json.

A LangChain service/module exists and exports at least:

Initialized LLM client.

One reusable function to run a basic AI prompt.

No hard-coded API keys or model names in code (must come from env).

Code is modular and can be used easily from controllers/routes.

Error handling is implemented for missing API key or failed AI calls.

💡 Nice-to-Have (Optional for this issue)

Logging for each AI call (prompt + timestamp).

Configurable timeout / max tokens / temperature.

Wrapper for future features like:

RAG (retriever + vector store).

System instruction injection based on DB content.