📘 Description
This project demonstrates a modular approach to building a multilingual translator using the LangChain framework and OpenAI’s GPT-4o-mini model. Designed for the MSDS 442 course at Northwestern University, the script highlights secure API integration, prompt engineering, and real-time language translation using large language models (LLMs).

More importantly, this project reinforces the idea that ChatGPT’s GUI is not designed to program business process workflows. Instead, we must create explicit reasoning and action workflows — like flowcharts or algorithms — to guide AI agents. This is why we leverage frameworks such as LangChain, LangGraph, or LlamaIndex, which act as bridges between our agent logic and the backend LLMs (e.g., OpenAI, Replicate, or Ollama).

📘 Description

This project builds on Module 1 by showcasing a modular agent framework using OpenAI’s GPT-4o-mini for multi-step problem-solving. Created for MSDS 442 at Northwestern, it follows a structured reasoning loop—Thought, Action, PAUSE, Observation—and integrates tools like LangChain for dynamic data retrieval and calculations. The project highlights how well-structured workflows enable AI agents to reason clearly and handle complex tasks beyond simple Q&A.

📘 Description

This project implements a functional memory-enabled chatbot using LangChain, OpenAI’s GPT-4o-mini, and LangGraph’s agent orchestration. It demonstrates how agents can reason and act in multiple steps, maintain session-based memory, and integrate with external tools for live information retrieval.

📘 Description

This project implements a dynamic, memory-enabled conversational agent that interacts over multiple turns, reasons through ReAct, and streams its thought process using LangGraph and GPT-4o-mini. Built within a Jupyter Notebook, the assistant simulates natural conversation with embedded search capabilities and message-based memory.

📘 Description

This project demonstrates how to build a basic semantic search engine over a real-world 10-K corporate filing. It serves as the foundation for more advanced retrieval-augmented generation (RAG) systems, showcasing how to move from unstructured PDF text to meaningful, queryable insights.

📘 Description

This project enhances the Nike 10-K semantic search pipeline by introducing structured metadata filters and query rewriting capabilities. It moves beyond simple similarity search by allowing users to ask targeted questions tied to specific sections of the document (e.g., “beginning”, “middle”, “end”).

📘 Description

This project extends a Retrieval-Augmented Generation (RAG) pipeline into a stateful, conversational assistant using LangChain, LangGraph, and OpenAI’s GPT-4o-mini. Designed for the MSDS 442 course at Northwestern University, it builds on previous modules by introducing agent reasoning, memory persistence, and multi-step retrieval workflows.

📘 Description

This project implements a financial planning agent using LangChain, LangGraph, and GPT-4o-mini. The assistant engages in natural conversations, interprets user intent, and helps with core financial tasks such as budgeting, savings goals, and net worth analysis. The system combines retrieval, prompt templating, and structured tool usage to provide intelligent, step-by-step financial guidance.

📘 Description

This project demonstrates a LangGraph-orchestrated Retrieval-Augmented Generation (RAG) system for intelligent querying of Nike’s 2023 10-K filing. It extends prior modules by introducing a graph-based control flow, allowing the app to analyze questions, retrieve relevant vector data using metadata filters, and generate structured responses — all as composable LangGraph nodes.

📘 Description

This module presents an AI-powered assistant designed to support hospital operations at Northwestern Memorial Hospital. Built with LangChain, LangGraph, and GPT-4o-mini, the agent handles multiple roles including triaging cases, analyzing operational metrics, and coordinating recommendations across specialist tools. It simulates a supervisory persona (“Claire”) that routes tasks to sub-agents and synthesizes responses in plain language.