Penmentor surfaced on Hacker News as a Show HN project that feeds handwritten math pages to Claude for teacher-style markup. The post collected 60 points and 7 comments within the first day.
The GitHub repository at https://github.com/AbdullahAlhariri/penmentor contains the core code and example images.
What It Is
Penmentor takes a photo or scan of handwritten math work and passes it to Claude’s vision model. Claude returns the original image overlaid with corrections, comments, and step-by-step guidance in the style of a human grader.
The tool focuses on one narrow task: turning static handwritten solutions into annotated feedback without requiring the user to retype equations.
How It Works
Users upload an image to a simple interface. The script extracts the image, constructs a prompt that instructs Claude to act as a math teacher, and returns the marked-up result. No fine-tuning or additional models are involved.
The workflow relies entirely on Claude’s existing vision and reasoning capabilities plus a lightweight wrapper script.
How to Try It
Clone the repository from https://github.com/AbdullahAlhariri/penmentor and install the listed Python dependencies. Add an Anthropic API key, then run the provided script with a local image path.
The README includes sample images and the exact prompt template used for grading.
Community Reception
Early HN commenters noted the narrow but practical scope. Several users asked about accuracy on messy handwriting and whether the tool handles multi-page proofs.
One thread highlighted the low barrier: anyone with a Claude API key can run it locally without additional training data.
Pros and Cons
- Pros: Requires only an existing API key; produces readable annotations directly on the image; runs on consumer hardware.
- Cons: Depends on Claude’s vision limits for very small or crossed-out writing; no offline mode; feedback quality varies with prompt phrasing.
Alternatives and Comparisons
Several established tools handle math feedback but differ in input method and output style.
| Tool | Input Type | Output Style | Offline Use | Cost Model |
|---|---|---|---|---|
| Penmentor | Handwritten photo | Image annotations | No | API per call |
| Photomath | Printed or typed | Step-by-step text | Partial | Freemium |
| GPT-4o vision | Any image | Text description | No | API per token |
| Wolfram Alpha | Typed expressions | Symbolic solution | No | Subscription |
Penmentor is the only option among these that returns corrections drawn directly onto the original handwritten page.
Who Should Use This
Students who want immediate visual feedback on paper homework will find it useful. Teachers testing automated grading workflows can use it for quick prototypes.
Users needing fully offline operation or guaranteed accuracy on complex proofs should skip it and keep human review.
Bottom line: Penmentor demonstrates a minimal, effective way to apply Claude’s vision model to handwritten math grading without building custom infrastructure.
The project shows how narrow, well-scoped wrappers around frontier vision models can deliver immediate classroom utility today.
Top comments (0)