Enterprise AI Course Agent with Persistent Memory & Token Tracking
An end-to-end, full-stack AI agent that automates curriculum design for instructional designers, featuring persistent conversational memory and automated API cost tracking.
Business Challenge
Instructional designers spent around 6 hours manually researching and structuring curriculum plans. Furthermore, integrating LLMs directly into the frontend posed security risks (API key exposure) and lacked a mechanism to accurately track and control the AI API budget (token usage) per user.
Solution Delivered
I architected a decoupled, stateful full-stack solution from 0 to 1. I built a robust backend using Python (FastAPI) to securely orchestrate the Google Gemini API. I implemented a SQLite database to maintain persistent session state and automatically calculate/log token usage. For the frontend, I developed a responsive, ChatGPT-like interface using Next.js and Tailwind CSS. Finally, I independently deployed the entire system on a private Linux VPS using Nginx as a reverse proxy (to resolve CORS issues) and PM2 for zero-downtime automation.
Outcomes Achieved
Reduced the manual curriculum planning process from 6 hours to under 2 minutes per course. Established a secure, production-ready infrastructure with 100% visibility into API costs per user session.
Measurable Business Outcome
Reduced manual curriculum design time by 99.4% (from 6 hours to under 2 minutes). Implemented 100% accurate token tracking for precise API budget control.