Data Science Tutorials

Can an AI agent reason like a scientist when the answer is hidden from it? Build a Python agent with Claude Fable 5.2 that tests competing explanations for noisy data, designs a follow-up experiment, and predicts results it has never seen, graded by separate code it can’t touch.

Use Copilot Cowork to turn one sales CSV into an Excel workbook, Word briefing, PowerPoint deck, and Outlook draft, then check the results, save the workflow as a skill, and schedule it.

What if your AI coding agent could improve itself? We tested Google's Harness^2 with OpenCode and DeepSeek on JobBench to find out whether self-improving AI agents actually perform better.

Learn how to run powerful decision models locally, explore the vLLM ecosystem, and build three practical projects using text, images, and videos.

Build a GitHub CI repair workflow with DeepSeek Superagent, then use observe mode, pytest, holdout data, and human review to check one pandas fix.

Build a Python agent with Gemini 4 Argon, function calling, and profiling, then check its patches with hidden tests, benchmarks, and API cost.

Learn how Mercury 3 uses diffusion-based generation to build an incident analysis assistant, visualize refinement states, and compare low vs. high reasoning effort.

Use Meta Muse Wonder and Olostep web API to check medical claims, explain the reasoning, and show where the evidence came from.

Build a Python image-to-CAD agent with the Kimi K3.1 Pro API using vision input, tool calling, a Docker sandbox, a hidden evaluator, and cost tracking.

Learn what Gemini Embedding 2 is, how it compares to Nova, Voyage, and Qwen, and build a multimodal search app in Python with ChromaDB.

A hands-on Meta Muse setup guide for first-time users. Learn how to set permissions, connect tools, scope a task the agent can actually finish, and read the Activity log so you know what it did before you trust the output.

An autoregressive model predicts future values of a time series from its own past. Learn how to fit one in Python with statsmodels, choose the lag order with PACF and AIC/BIC, and compare it to MA and ARIMA.