Project ideas from Hacker News discussions.

Students prefer Gemini over ChatGPT and Claude for AI essays in blind tests

📝 Discussion Summary (Click to expand)

Theme 1 – Gemini excels at natural‑language writing and clear explanations
- “Gemini is bad at coding, but good at writing. More human like” – ShawWang
- “Gemini excels at explaining new concepts clearly and intuitively. Much more so than other models.” – John7878781
- “Gemini is my last pick, but does better than the rest for making something clear and understandable.” – gexla
- “Gemini simply hallucinates less … important for students.” – dzhiurgis

Theme 2 – Claude’s prose style is often seen as painful, bureaucratic, or “Claudish”
- “The 'Claudish' has become so painful to read that it just takes me out of whatever task I was working on.” – StableAlkyne
- “Claude speaks Claudish … its prose has become painful.” – gherkinnn
- “Claude's 'English' is very obviously unnatural‑sounding, and extremely distinctive in a bad and irritating way.” – userbinator
- “When prompted to use simple English without jargon, it's still filled with load bearing honest caveats in every footgun seam.” – StableAlkyne

Theme 3 – Developers tend to split tasks by model strength (coding vs. writing/comments)
- “I use Qwen/Claude for coding and Gemini/Gemma to draft text intended for a human. To me that feels like each model playing to their strengths.” – SturgeonsLaw
- “You could just switch the model after the code is written, when you need to ask questions.” – pkulak
- “I use Gemini for stray conversations… Gemini has a better conversational style than GPT… I use Gemini for the ‘everything else’ category.” – adventured
- “Has anyone tried letting Claude or GPT do the coding, and Gemini do the comments and documentation?” – inbx0


🚀 Project Ideas

SplitMind

Summary

  • Routes code generation prompts to coding‑optimized LLMs (Claude, Qwen, DeepSeek) and natural‑language tasks (comments, docs, essays) to writing‑optimized LLMs (Gemini, GPT‑4).
  • Core value: seamless, high‑quality code and clear documentation without manual model switching.

Details

Key Value
Target Audience Software developers, devops, data scientists using LLM‑assisted coding
Core Feature Intelligent task classifier + model router that sends each prompt to the best‑suited backend and returns a unified response
Tech Stack Python/FastAPI backend, lightweight classifier (fine‑tuned BERT or zero‑shot), APIs for Anthropic, Google Gemini, DeepSeek, VS Code extension or CLI
Difficulty Medium
Monetization Revenue-ready: Subscription tiered by API calls (e.g., $10/mo for 500k tokens)

Notes

  • HN commenters said: “I use Qwen/Claude for coding and Gemini/Gemma to draft text intended for a human.” – SturgeonsLaw
  • Also: “makes sense to let Claude or GPT do the coding, and Gemini do the comments and documentation” – inbx0
  • Potential for discussion: reduces token waste, improves workflow, open to custom model routing strategies.

PlainClaude

Summary

  • Takes verbose, jargon‑heavy Claude output (comments, error messages, explanations) and rewrites it into concise, natural language.
  • Core value: improves readability of AI‑generated text, lowering cognitive load when reviewing LLM‑produced docs.

Details

Key Value
Target Audience Developers, technical writers, students who frequently read LLM‑generated comments or documentation
Core Feature Simplification pipeline: input Claude text → summarization/paraphrasing model (Gemini or fine‑tuned T5) → plain English output with optional length control
Tech Stack VS Code extension or web UI; Python/FastAPI backend calling Gemini API or open‑source model (e.g., Flan‑T5); optional local model for privacy
Difficulty Low‑Medium
Monetization Hobby

Notes

  • HN commenters noted: “Claude speaks Claudish … I have to read a lot of prose and I simply cannot get myself to read a page of claudisms.” – gherkinnn
  • Also: “Claude speaks Claudish but at the end of the day I care about the Ruby, Python, JS it writes…” – pmontra (shows desire for better prose)
  • Potential: could be integrated into CI to auto‑clean generated documentation, reducing noise and improving team productivity.

EssayLens

Summary

  • Provides length‑normalized essay evaluation, focusing on argument quality, clarity, and originality rather than verbose length.
  • Core value: helps educators grade fairly and helps students improve writing without gaming length.

Details

Key Value
Target Audience Teachers, professors, students (high school/college) using AI for essays
Core Feature Rubric‑based scoring (thesis, evidence, organization, style) + AI feedback that highlights strengths/weaknesses; includes AI‑generated text detection and length‑bias correction
Tech Stack Web app (React/Next.js) + Python/FastAPI backend using Gemini or GPT‑4 for feedback, plus lightweight AI‑detector (RoBERTa‑based)
Difficulty Medium
Monetization Revenue-ready: Subscription per institution or per student (e.g., $5/student/month)

Notes

  • HN commenters observed: “Students tended to prefer longer responses. The selected answer was 37% longer on average than the alternatives.” – pasharayan
  • Also: “I'd love to see the same experiment with responses normalized for length, and with actual essay quality scored separately from how helpful the model's feedback felt” – Honali
  • Potential: addresses frustration about AI essay bias, could spark discussion on fair grading in the age of LLMs.

Read Later