IdeaWell

Project ideas from Hacker News discussions.

Vomit: Clean up Claude 5's token output with a separate LLM

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Verbose “Claudish” output

"The baked in communication style of these models is so obnoxious it’s impacting my work." — Bluestein

2. Tools to tame the style

"I added a skill I call “deslop” to clean up Claude output when it goes off the rails." — bmurphy1976

3. Model degradation affecting mental health

"My quality of life ... improved drastically since I stopped subjecting myself to reading this style of output 8 hours a day." — ericpauley

4. Community work‑arounds and rewrites

"I think this needs a before and after example." — pebbly_bread

🚀 Project Ideas

Generating project ideas…

Claudish-to-English API

Summary

  • Convert verbose LLM output into clear, conversational prose.
  • Preserve all factual details while trimming filler.

Details

Key Value
Target Audience Developers, technical writers, product managers who need concise AI-generated text.
Core Feature REST API that rewrites any prompt output into plain English, with optional style presets.
Tech Stack FastAPI backend, GPT‑3.5‑turbo for rewriting, Redis cache.
Difficulty Medium
Monetization Revenue-ready: $0.02 per 1k tokens

Notes

  • Users like alansaber said “I’d love to know what the hell Anthropic has done to make Claude’s writing so… bad.” and pickledish liked the example.
  • Directly solves the “Claudish to English” pain point repeatedly mentioned in the discussion.

Style‑Enforcer Hook Suite for LLM Agents

Summary

  • Provides a ready‑to‑use hook library that injects concise output rules on every turn.
  • Monitors and trims excessive jargon automatically.

Details

Key Value
Target Audience Engineers building agentic workflows with Claude Code, GitHub Copilot X, or similar.
Core Feature CLI plugin that adds a persistent style‑preservation block to prompts, auto‑detects verbose patterns, and rewrites on‑the‑fly.
Tech Stack Python package, OpenAI‑compatible token limits, Docker container for sandboxed execution.
Difficulty Low
Monetization Hobby

Notes

  • Discussion highlights “hooks need to be repeated every turn” (Bluestein) and “waste tons of tokens (repeated verbatim…)” (zachahn).
  • Directly addresses the token‑waste and style‑drift complaints.

Plain‑English Code Review Summarizer

Summary

  • Generates succinct PR description and comment summaries from verbose AI explanations.
  • Highlights only changed code and rationale.

Details

Key Value
Target Audience Teams using AI‑assisted code review (e.g., Anthropic’s Claude Code).
Core Feature CLI tool that takes a diff and AI‑generated commentary, outputs a 3‑bullet summary in plain language.
Tech Stack Node.js, GPT‑4o‑mini for summarization, Markdown output.
Difficulty Medium
Monetization Revenue-ready: $9/mo per user

Notes

  • Commenters like alansaber noted “I’d love to know what the hell Anthropic has done…” and cjk mentioned summarizing works “remarkably well”.
  • Solves the “excessive verbose comments” issue highlighted throughout the thread.

Personalized Output‑Style Coach (SLACK‑Integrated)

Summary

  • Learns a user’s preferred concise writing style and applies it to all AI interactions.
  • Provides real‑time feedback on overly verbose responses.

Details

Key Value
Target Audience Power users of LLMs who need consistent, readable output across platforms.
Core Feature SaaS that syncs with Slack/Teams and rewrites AI messages on the fly, storing style preferences via user feedback.
Tech Stack React frontend, FastAPI backend, fine‑tuned Llama‑3 8B for style enforcement.
Difficulty High
Monetization Revenue-ready: $30/mo per seat

Notes

  • Users express desire for “a way to reliably change the way an LLM responds” (Bluestein) and “the baked‑in communication style … is so obnoxious”.
  • Tackles the systemic style problem discussed across many comments.

Read Later