Project ideas from Hacker News discussions.

Don't Use AI to Write

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Generating summary…

🚀 Project Ideas

Generating project ideas…

DraftMate

Summary

Provides a side‑by‑side diff editor that shows original text and AI‑suggested edits, forcing the user to review and accept/reject each change.
Encourages disciplined use of LLMs by treating AI output as a draft that must be edited, preserving the writer’s voice and effort.

Details

Key	Value
Target Audience	Writers, professionals, non‑native speakers who use LLMs for drafting
Core Feature	Interactive diff editor with mandatory review workflow, AI‑ism highlighting, and edit‑accept/reject controls
Tech Stack	Electron (React + Monaco Editor) frontend, Node.js/FastAPI backend for LLM calls (OpenAI, Ollama), optional local model support
Difficulty	Medium
Monetization	Revenue-ready: Subscription $5/mo per user (team plans available)
#### Notes
- HN commenters complained about “one‑shot” AI output and losing the thinking process; DraftMate makes the AI a collaborator that must be edited, preserving the writer’s voice and effort.
- Provides a tangible way to practice the discipline praised by users who treat LLMs like a word‑processor—users see exactly what changed and can iterate, turning AI assistance into a learning tool rather than a crutch.

MeaningCheck

Summary

Generates a forward translation, then automatically back‑translates to the source language and scores semantic divergence.
Highlights segments where meaning may have drifted, giving users confidence in AI‑generated translations.

Details

Key	Value
Target Audience	Translators, multilingual professionals, language learners who need confidence in AI‑generated translations
Core Feature	Forward translation + automatic back‑translation with similarity scoring and drift highlights
Tech Stack	Web app (React/Vue), translation APIs (DeepL, Google, or open‑source models), sentence‑transformer model for similarity scoring, hosted on Docker
Difficulty	Medium
Monetization	Revenue-ready: Pay‑per‑character API ($0.0005 per 1k chars) with a free tier of 10k chars/month
#### Notes
- Users warned against trusting AI translation without verification (sasaf5: “don’t use AI to translate into a language you can’t read back”); MeaningCheck gives an automatic back‑translation check so you can read the output in your own language.
- By exposing meaning loss, the tool encourages the careful, elbow‑grease approach that HN commenters advocate for using LLMs as a second‑pass rather than a final product.

StyleGuard

Summary

Real‑time detector of common LLM‑isms, filler phrases, and potential hallucinations in any text.
Offers rewrite suggestions and tracks AI‑usage statistics to help writers keep AI output from slipping into final work.

Details

Key	Value
Target Audience	Editors, writers, academics, anyone who wants to keep AI‑generated text from slipping into their work
Core Feature	Real‑time detection of AI‑isms and hallucinations, rewrite suggestions, usage analytics
Tech Stack	Browser extension (JS) + lightweight backend (Python/FastAPI) using regex rules + a small classifier (e.g., DistilBERT) hosted on a cheap VPS or serverless
Difficulty	Low
Monetization	Hobby (open‑source) – can be extended with premium rule sets for $3/mo if desired
#### Notes
- Commenters complained about “LLM‑isms like ‘not this, but that’” adding zero value and signaling low effort; StyleGuard flags those patterns so writers can strip them out before publishing.
- Provides the practical utility of a proofreading second pass that users like nikbou and basilikum praised, while encouraging the mindset of treating AI output as a draft to be edited, not as final copy.