LangChain - Develop Controlled AI Agent with LangChain & RAG

Demonstrates a business ai agent built with LangChain and a vector db using RAG to generate a business plan, marketing strategy, emails, and tasks from user context.

Select a good code editor for the project, with Visual Studio Code recommended; you may use others like Sublime, Atom, or Cursor, and Windows users may install Git Bash.

Access the official course git repo to follow along, clone code from eight branches—planning and marketing chain, email chain, task chain, ad reviewer—with the final code in branch eight.

Initialize the LangChain ai agent project with fast api and ubicone, set up virtual environment, install packages, create main.py, and outline folders for agents, api, chains, prompts, schemas, and ui.

Create a config.py to load the root .env file, validate the OpenAI API key, and set the model name and base and prompts directories.

Acquire an OpenAI API key, create a project, and generate a secret key; deposit credits, review pricing, and paste the key into your env file.

Configure the app router and UI to run the AI agent, mounting /app, serving a Tailwind-based HTML UI and a FastAPI post route for run agent with form data.

Design the business and system prompts for a LangChain AI agent. Outline a JSON-output workflow with schemas, workers, and a manager for business tasks.

Define a business overview schema using pydantic to enforce a five-field, structured output. Ensure consistent prompts and validation for summary, audience, core pain point, unique value proposition, and priority note.

Learn to build a business chain with LangChain, integrating system and business prompts, initializing an LLM, and structuring output via a chained workflow for consistent AI task execution.

Create an executor agent that streams real-time thoughts and logs while running a business chain, using tone and depth settings and server-sent events.

Learn to run a LangChain agent from a front-end, sending task, tone, and depth as form data, streaming real-time logs and results to render a live summary.

Develop a planner prompt to convert business tasks into an executable plan and output a JSON steps list including business analysis, marketing strategy, email campaign, and task breakdown.

Develop a planner agent by defining an execution plan schema, configuring a deterministic LLM-driven planner with tone and depth, and generating actionable steps via a planner chain and real-time logs.

Learn to implement agent logs with a create_log function, plan and execute chained steps, and stream real-time log content for business analysis, email campaigns, and marketing tasks.

Develop a practical marketing prompt and schema to generate a json-based marketing strategy, detailing primary goals, core messages, prioritized channels, and ignored channels.

Build and run a marketing chain with an LLM to generate a marketing strategy that respects tone and brand voice, render results on the front end, and validate final outputs.

Learn to craft an email prompt and a JSON output schema to generate three sequential emails with subject, objective, body, and call to action for a LangChain workflow.

Build an emails chain in LangChain to draft multiple professional emails using prompts and a structured email list. Integrate and test the chain within the app.

Create a task prompt and a task schema, then plan a task chain. Export a json-formatted list of tasks with order, name, description, priority, and why it matters for business.

Develop a task chain in LangChain to break down a business task into prioritized tasks, using prompts, a low-temperature llm, and a reviewer and ranking workflow.

Learn how to build a reviewer agent by defining a system and user prompt to review the provided JSON business plan before delivery, ensuring tone and completeness.

Develop a reviewer chain to evaluate and improve model outputs using prompts and json formatting. Validate responses through a reviewer prompt and system prompt, applying quality assurance and improvements.