Project Ideas
Project ideas and real projects from people who've taken courses. End-to-end AI applications and agentic workflows you can learn from and build on.
Carlos Pumar-Frohberg's agent analyzes client satisfaction using Stack Exchange data, focusing on UI discussions to find frustration patterns. An orchestrator agent routes questions to a MongoDB agent for 'what'/'how' queries or a Cipher agent that translates natural language into Neo4j graph queries, often calling both for safety.
A reference project from the AI Engineering Buildcamp: an agent that interacts with a simple to-do list application. Built with Lovable for the frontend and FastAPI (Python) for the backend. Uses the backend's OpenAPI spec so the model can create tools to get tasks or mark them complete, with Logfire for monitoring and pytest for testing.
A small, self-contained project idea: take a photo of an everyday object and turn it into a short story with AI. Expand gradually with illustration, audio, a small website, and a podcast feed. Great for experimenting with multimodal pipelines (vision, text, speech) and sharing with kids, family, or friends.
A full-stack hospital operations platform (React + Express + MongoDB) for patient registry, appointments, medical records, and ward occupancy. Reference implementation for sensitive-data systems: role-based access control, audit logging, input validation, rate limiting, and soft-delete patterns built in from the start.
Asia Amodeo's intelligent email agent helps manage an inbox. It integrates with the Gmail API, indexes emails in Elasticsearch, and offers a Streamlit chat interface to find specific emails or see what's important for the day, addressing 'email fatigue' by making it easier to sift through communication.
Xia He-Bleinagel built a full-stack holiday planning app on Django using AI tools: ChatGPT for prompts, Lovable for UI, Cursor for implementation, and Google Antigravity (Gemini) for debugging. Features Supabase auth, role-based access, and integrated school and German national holidays. No more static spreadsheets.
Vancesca Dinh's agent helps users identify goals and understand the 'why' behind them, grounded in data from the Huberman Lab podcast and medical publications. Pipeline includes RSS feeds, Faster Whisper transcription, Qdrant embeddings, query rewriting for better search, Logfire, Pydantic, and guardrails to keep the agent on-task.
An agent that tracks cybersecurity incidents reported to the SEC, built by Scott DeGeest. Helps supply chain professionals quickly find out if a company has disclosed a data breach or ransomware attack. Handles raw XML/PDF from the SEC, invalid XML, subsidiary relationships (e.g. Change Healthcare ↔ United Health Group), and token monitoring for cost and context limits.
A production-style app that automates messy CSV cleanup: upload a file, profile it, generate AI-backed cleaning steps, apply deterministic transformations, and download a cleaned dataset with a human-readable report. Turns a repetitive, error-prone task into a repeatable pipeline with tests, an API, and Cloud Run deployment. LLM acts as planner, cleaning engine is deterministic for reliability.