🤖 DS205 AI Tutor

I have set up a custom Claude for our course. Students in DS105 found it genuinely useful, and the DS205 version knows even more about what you’re building.

Getting Access

Follow these instructions to get started:

1. Make sure you have access to the LSE Claude Enterprise account.
   • Fill out this form to get access.
   • Click here to learn more about LSE’s partnership with Anthropic for the Claude for Education account.
2. Give me your email address: Go to the #announcements channel on Slack and look for my pinned message so I can add you to the list of users who have access to the DS205 Claude tutor.

What Makes This Different

The DS205 Claude tutor knows our course. It understands the weekly structure, what you’ve covered so far, and what’s coming next. When you ask a question, it responds within the boundaries of where you actually are in the course.

Figure 1: The tutor knows that Problem Set 1 instructions release after Tuesday’s lab, not at the start of W02. It tells you what to focus on now rather than leaving you confused.

Generic AI tools have no sense of timing. They might explain something using concepts from Week 08 when you’re still in Week 02, or reference materials you haven’t seen yet. Our tutor stays grounded in what’s been covered.

Code Review That Asks Questions

When you share code for feedback, the tutor validates that it works, then asks you to think through the next steps rather than rewriting your code for you.

Figure 2: The tutor acknowledges what’s working (the connection, the selector choices), then asks architectural questions: where does the data go? What happens when something’s missing? These are the questions you’ll need to answer for your Problem Set.

You’ll notice it points out the time.sleep(5) approach without demanding you change it immediately. The goal is to help you understand trade-offs, not to produce perfect code on the first attempt.

Debugging With You, Not For You

When something isn’t working, the tutor explains what’s likely happening and gives you a diagnostic technique to investigate further.

Figure 3: The selector works for some products but fails for others. The tutor identifies the key clue in the pattern, asks diagnostic questions about what element and what the HTML looks like, and gives a concrete technique: manually inspect products that fail and compare them to ones that work.

This matters because web scraping problems are rarely identical. The tutor helps you build a mental model for diagnosing issues yourself, which is more useful than a one-off fix.

Curiosity Gets Encouraged, Then Redirected

If you ask about topics from later in the course, the tutor will explain them accessibly, then connect back to what you’re doing now.

Figure 4: When asked about embeddings and RAG systems (W07-W11 content), the tutor gives a genuine explanation rather than refusing to engage. But it also maps the course progression and explains why current scraping skills are foundational.

You’re not blocked from exploring ahead. But the tutor helps you see how current work enables later work, so you understand why the foundations matter.

It Complements the Course

Think of the tutor as an extension of office hours. It can help you understand what a task is asking, clarify terminology, talk through your approach, or point you back to relevant lecture material. It cannot attend lectures for you or replace working through the notebooks yourself.

The goal is to help you engage more deeply with the course, not to create a shortcut around it.

Warning

Chatbots will occasionally give inaccurate advice. The technology behind generative AI does not have a concept of “true”; it generates whatever sounds most plausible. This tutor is no exception. I’ve configured it to align with the course, but it won’t always be right. Use your judgement.