Inspiration
txt2crs started with a simple question from my wife: “How can we bring AI to Africa?” It reminded me of an earlier course-generation solution I built and inspired me to turn that idea into a complete learning tool—one that can make structured educational material easier to create and access.
What it does
txt2crs turns a topic or source into a complete, research-backed learning package. Users can submit text, a webpage, a YouTube link, or a PDF, DOCX, or PPTX file.
From that single input, txt2crs creates:
- A structured course
- A comprehensive review pack
- A student assessment
- A separate instructor answer key
Each resource is available in HTML, Markdown, PDF, and DOCX formats, with sources and conflicting information clearly disclosed.
How we built it
We built txt2crs with a React frontend, a FastAPI backend, and a reusable Python generation engine. GPT-5.6-SOL handles the course creation, while Tavily provides focused research through a restricted MCP server.
The generation process is split into clear stages: input validation, research, course planning, lesson writing, assessment creation, quality checks, and document rendering. Jobs are saved as they progress, so users can refresh the page or restart the application without losing their work.
Challenges we ran into
The biggest challenge was making AI-generated content dependable rather than simply impressive. We needed to verify citations, catch unsupported claims, keep the course and assessment aligned, and recover safely when a generation step failed.
We also had to support several input and document formats while keeping every user’s files private and the experience easy to understand.
Accomplishments that we're proud of
We are proud that txt2crs produces four connected learning resources instead of a single block of generated text. The course, review material, assessment, and answer key are built from the same validated content, making the final package more consistent and useful.
We are also proud of the durable job system, source-grounded research, private file delivery, accessible interface, and strong automated test coverage.
What we learned
We learned that good educational AI needs more than a strong prompt. It needs clear stages, trusted research, strict validation, thoughtful failure handling, and a user experience that explains what is happening.
We also learned that model output should be treated as a draft candidate— the application must remain responsible for deciding whether that output is complete, supported, and safe.
What's next for txt2crs
Next, we want to add course editing, more languages, richer media support, LMS exports, collaboration tools, and automatic grading. Our long-term goal is to make high-quality learning materials easier to create for educators and learners anywhere.
Log in or sign up for Devpost to join the conversation.