AI-Powered Syllabus Alignment for Calculus Textbooks

Aligning a calculus textbook with AP syllabus requirements is a slow, manual process. In an experiment with the Calculus Consortium, we used AI agents and semantic search to accelerate this work. We found an expert, aided by an AI agent with search tools, can complete alignment tasks dramatically faster.

We started with search.

Search Tooling

We indexed the first ten chapters of a calculus textbook using Mixedbread. This took ~ ten minutes and cost $30, creating a "superindex" of its content.

Searching the "superindex" with the Mixedbread Cloud interface helped locate key sections, but the process remained slow and manual.

From Search Tools to AI Agents

Encouraged by the retrieval results, we explored two workflows:

• A custom web application built with FastAPI to provide a search interface.
• General AI Agent with a tool to search the Mixedbread index.

Custom Web Application

We started with a workflow-driven web application we built. While promising, it lacked flexibility. This approach may be useful after several alignments when core patterns are clear.

Open Hands And Claude Code

We gave the AI agent the AP syllabus and access to the indexed textbook. The autonomous agent workflow proved most effective. The reviewing mathematician called the report on 15 syllabus items:

"Amazing, brilliant even. It found conceptually related material, not just keyword matches. As if it understood the math it was searching for. I did have to go through the report item by item to verify accuracy."

The agent provided a head start, locating relevant page numbers, exercises, and problems for each syllabus item. Open Hands proved most effective because of the cloud UX.

To guide the successful agent, we gave it a simple set of instructions via an AGENTS.md file:

Loading...

The agent then used the mxbai command-line tool to query the index, analyze the results, and generate detailed reports for each curriculum item.

The Results: Actionable Alignment Reports

The agent produced a structured, actionable Markdown report for each syllabus item.

For well-covered items, the report confirmed coverage with evidence:

Loading...

The agent also synthesized these individual files into a single overall summary report, giving the author a high-level view of the work the agent felt was required to bring the book into full alignment.

Conclusion

Mixedbread's semantic search and an autonomous AI agent proved extremely effective for curriculum alignment.

• Semantic search found conceptually related material, not just keyword matches.
• The AI agent automated the complex workflow of searching, synthesizing, and reporting.

Coding dominates the AI agent conversation. But the pattern here applies across industries: experts bottlenecked by tedious search and synthesis.