AI Practice & Projects

Call an LLM directly from Python — beyond the chat interface. Send a prompt, parse the response, and handle errors and retries.

Run a pretrained sentiment model over a batch of reviews or tweets and visualize the results — your first real ML pipeline, start to finish.

Use a pretrained vision model to classify your own set of images, then build a tiny web page to upload and label new ones.

Build a basic content-based recommender for movies or books using similarity scores over a small public dataset.

Index a small document set into a vector store, wire up retrieval, and prompt an LLM to answer questions grounded in that content.

Use LoRA or another parameter-efficient method to fine-tune a small open model on a narrow task, like tone or a specific format.

Build a pipeline that flags text for policy violations using a moderation model, with a human-review queue for uncertain cases.

Build a chat app that remembers what a learner already knows across sessions and adjusts explanations to their level.

Coordinate multiple agents with different roles — search, summarize, critique — to research a topic and produce a cited report.

Train, version, deploy, and monitor a model with MLflow and Docker, including automatic retraining when performance drifts.

Build a test suite that scores an LLM app's outputs against a labeled dataset — accuracy, faithfulness, and refusal rate.

Chain speech-to-text, an LLM, and text-to-speech into a low-latency voice assistant that can hold a real conversation.