š Project Ideas
Generating project ideas…
Summary
- Browser extension that adds a nuanced "I regret watching this" button (with optional sentiment tags) to capture negative feedback that current like/dislike buttons miss.
- Core value: gives users agency to tell recommendation algorithms they engaged reluctantly, reducing unwanted content spirals.
Details
| Key |
Value |
| Target Audience |
Power users of YouTube, Netflix, Spotify who feel trapped by recommendation loops |
| Core Feature |
Oneāclick regret button + sentiment tagging that feeds a local ML model to downārank similar content |
| Tech Stack |
TypeScript, React, WebExtensions API, TensorFlow.js for onādevice inference |
| Difficulty |
Medium |
| Monetization |
Hobby |
Notes
- HN users complained āIām engaging with this, but I hate myself for doing itā (bryanrasmussen) and wanted a way to signal regret (graemep). This directly addresses that.
- Could evolve into a crossāplatform feedback standard, sparking discussion about algorithmic accountability.
Summary
- A personal proxy/intermediary that rewrites recommendation feeds from platforms according to userādefined rules (topic caps, blocklists, interest decay).
- Core value: lets users enforce diversity limits (āno more than 5% about bad economyā) and break echo chambers without leaving the platform.
Details
| Key |
Value |
| Target Audience |
Users who want control over their algorithmic diet (e.g., news, video, music consumers) |
| Core Feature |
Rules engine that intercepts API responses or scraped feeds, applies topic classification and enforces caps/filters |
| Tech Stack |
Go (backend), Python (topic classification with spaCy/HuggingFace), Docker, optional browser extension for UI |
| Difficulty |
High |
| Monetization |
Revenue-ready: subscription $4/mo |
Notes
- HN commenters asked for āintermediary feeds ā¦ that lets us customize how much we get of each type of contentā (zug_zug). This idea implements that.
- Provides concrete utility for discussion on algorithmic transparency and user agency.
Summary
- An openāsource, localāfirst recommendation engine that learns from explicit user signals (likes, dislikes, regret, watch time) and runs entirely on the userās device.
- Core value: privacyāpreserving personalization that never sends raw behavior data to servers, while still delivering tailored suggestions.
Details
| Key |
Value |
| Target Audience |
Privacyāconscious users and developers who want to selfāhost recommendation logic |
| Core Feature |
Onādevice model (e.g., lightweight neural collaborative filtering) updated via user feedback; exportable to platform APIs via userācontrolled tokens |
| Tech Stack |
Rust (core), PyTorch Mobile or ONNX Runtime for inference, SQLite for local storage, optional WebAssembly wrapper for browser use |
| Difficulty |
High |
| Monetization |
Hobby |
Notes
- Users expressed distrust that feedback is used primarily to keep them on the platform (jjulius) and wanted alternatives that donāt rely on opaque servers (BuyMyBitcoins). OpenRec gives them full control.
- Could become a building block for many apps, fostering discussion about decentralized recommendation ecosystems.
