Proposal 2026: AI Powered Toxic Content or Spam Detection

I’m interested in working on this project idea for GSoC 2026 and would love to contribute. I’ll start reviewing the related modules and existing discussions to better understand the scope.

Please let me know if there are any specific directions or expectations I should keep in mind while preparing my proposal.

Hi @pooja_sharma,

I’m Akash (codeguyakash), a GSoC 2026 applicant interested in the AI-Powered Toxic Content and Spam Detection project.

I’ve started exploring possible approaches for this module and had a few thoughts:

- Implementing a modular AI integration layer (Detoxify / Hugging Face)
- Supporting both client-side (real-time feedback) and server-side validation
- Providing configurable thresholds (toxicity/spam) for admins
- Extending support across comments, node forms, and webforms

I’m currently setting up a local Drupal environment and will start working on a small prototype / POC for toxicity detection.

Would love your guidance on:
1. Preferred AI provider (Detoxify vs Hugging Face vs custom FastAPI)
2. Whether we should prioritize client-side detection or backend validation first
3. Any existing modules or prior work I should explore

Looking forward to contributing!

Thanks,
Akash

Hi @pooja_sharma,

I’m interested in the “AI Powered Toxic Content or Spam Detection” project and have been exploring how such a system can be effectively integrated within Drupal’s content workflows.

I had a few questions regarding the architecture and scope:

1. For model integration, is the expectation to rely primarily on external APIs (e.g., Hugging Face/Detoxify), or would hosting a lightweight model via a FastAPI service be preferred for better control and latency?

2. For real-time validation (client-side), should the detection be synchronous with API calls, or would a debounced/asynchronous approach be better to balance UX and performance?

3. How should false positives/negatives be handled? Is there interest in a configurable threshold system or admin feedback loop to improve moderation accuracy over time?

4. Should the module integrate with Drupal’s existing content moderation workflows (e.g., flagging content instead of blocking submission)?

5. For scalability, especially on high-traffic sites, would queue-based processing be expected for server-side validation?

I’m currently reviewing existing Drupal modules related to spam prevention and moderation, and experimenting with API-based toxicity detection approaches.

Any guidance on preferred direction or initial contribution areas would be very helpful.

Thanks!