I built a tool that filters AI slop out of English social posts. The hardest part was teaching AI to stop sounding like AI.
I'm a Chinese indie developer. I have ideas. My English is decent but not native. Every time I tried to post on X or LinkedIn, the result sounded like a textbook. Or worse, like ChatGPT.
The ironic part: when I asked ChatGPT to "make it sound natural," it added em dashes everywhere. You know the type. "I learned three things, and the third one surprised me." That's not how real people write. But AI keeps doing it because its training data is full of that pattern.
So I built VoiceBridge. You paste a Chinese thought. It rewrites it for X, LinkedIn, and Reddit in three completely different voices. Not translation. Cultural adaptation.
The technical problem was harder than I expected.
I'm using DeepSeek V3 for generation. First version was garbage. The output had em dashes everywhere, "Let me tell you" openings, "The lesson here is" endings, "Here's the thing" transitions. You get the idea.
So I built a three-layer defense system.
Layer 1: System prompt hard constraints.
Anti-fabrication rules. The AI is not allowed to invent user counts, ratings, tech stacks, timelines, dollar amounts, or team sizes. It also can't do the coaching ending thing. No "Sometimes the answer is simple." No "The real takeaway is." Just the content.
Layer 2: Cultural translation table.
This was a rabbit hole. "Neijuan" is not "involution." Nobody says involution. It's "rat race" or "burnout culture." "Chuhai" is not "going to sea." It's "going global." I built a lookup table of 30+ Chinese internet terms with their actual English equivalents. Not what Google Translate says. What native speakers actually type.
Layer 3: Server-side regex cleanup.
Because you cannot trust an LLM to follow instructions 100% of the time. The API does a final pass. Strips all em dashes. Flags 30+ AI-tell patterns with regex. Scores the output 0-100 on "AI-ness." If score is below 85, it gets re-generated.
The scoring was surprisingly effective. I tested it on posts from popular AI writing tools. Most scored 40-60. VoiceBridge outputs now consistently score 95-100.
Platform-specific voice.
X gets punchy. Short sentences. Contractions. No fluff.
LinkedIn gets longer form but not corporate-speak. Think a senior engineer writing a blog post, not a marketing team.
Reddit gets the humble version. "Hey folks, I built this thing, here's what happened." No hype words.
Each platform also gets structural recommendations. Reddit output includes which subreddit to post in and the best time to post.
The stack.
Next.js 16, TypeScript, Tailwind v4, DeepSeek V3 via SiliconFlow API, deployed on Vercel. The whole tool runs as a single API route with streaming. No database, no user accounts. Just paste and go.
Where it's at now.
Live at muzi.studio/tools/voicebridge. Free to use. No signup. I shipped it in about 2 weeks of evenings while my kids were asleep.
The part I'm most proud of is the AI score system. Seeing that "100/100 clean" badge on every output after months of reading AI-slop on social media is deeply satisfying.
What I'm struggling with.
Marketing. I'm a developer. I can build things. Getting people to actually try it is a completely different skill. I have 4 followers on X. The tool is free and needs no signup, but somehow that makes it harder to sell because people don't trust free.
What would make you trust and try a free AI writing tool?
If you have 2 minutes, I'd love a brutal roast of the landing page: muzi.studio/tools/voicebridge
Spot on about the two-problem split. Structural tells are the easy part, which is why I focused there first. Voice injection is the harder problem and you framed it perfectly. Right now VoiceBridge takes a raw thought plus platform context. I am experimenting with a hot-take field where you write one opinionated sentence and the model builds around it. Early results are promising. Thanks for genuinely helpful framing.
The 'stop sounding like AI' problem is actually two separate problems:
1. Structural tells - em dashes, unnecessary hedging, bullet points for everything. Filterable.
2. Voice absence - no specific opinion, no concrete story, no sharp edge. Not filterable because filtering cannot add what was never there.
The second problem is why most AI content feels hollow even after you strip the verbal tics. The words are fine but there is no perspective behind them.
The writers I have seen use AI well do not fight this - they front-load their own specific take, then use AI to build out the structure. 'Here is my actual opinion on X: [specific, opinionated take]. Now help me write a post that builds the argument.' That forces the model to work inside someone's worldview rather than defaulting to the median internet opinion.
The filtering tool addresses the structural tells well. Curious whether you have any hooks for injecting voice/perspective, or if that is intentionally out of scope.
Thanks for the sharp feedback on positioning. You are right that cultural adaptation is the real wedge, not AI writing. I have been going back and forth on whether to lead with the language angle or the AI-slop angle. Your comment makes me think I should lead with both: non-native voice crossing language barriers AND platform-specific AI cleanup. On the name, good point. VoiceBridge works for now but I can see how it sounds like a translation tool. Will think about that as the product evolves.
The strongest part here is “cultural adaptation,” not AI writing. Most writing tools promise better posts, but your wedge is much sharper: helping non-native founders turn real thoughts into platform-native English without sounding translated or AI-generated.
The three-layer system is also a good trust signal. Anti-fabrication rules, cultural phrase mapping, and AI-pattern cleanup make this feel more serious than a generic rewrite tool. I’d make that the core positioning: not “write better social posts,” but “keep your real voice while crossing language and platform barriers.”
One thing I’d watch is the VoiceBridge name. It explains the translation/voice idea, but it may still sound like a language tool. If this becomes a broader trust layer for founder communication, content, and global positioning, Beryxa .com would give it a cleaner SaaS-style brand than a feature-like name tied only to voice bridging.