Curious what made you think the backend uses LLMs for content generation?
To clarify:
1. transcription is local VOSK speech-to-text via WebSocket
2. live transcript post-processing has optional Gemini Flash-lite turned on which tries to fix obvious transcription mistakes, nothing else. The real fix here is more accurate transcriber.
3. backend: TypeGraphQL + MongoDB + Redis
The anti-AI stance isn't "zero AI anywhere", it's about requiring human input.
AI-generated audio is either too bad or too perfect. Real recorded voice has human imperfections.
I'm working on https://X11.Social, a voice-first content creation tool for X.
The initial idea was "call to tweet", the ability to compose posts on the go by having a natural conversation with an AI assistant over a simple phone call. This is useful for turning thoughts from a walk or drive into a polished "brain dump" post, or for engaging with user lists without being at a computer.
It has since evolved into a broader system:
Chrome Extension: A context-aware assistant that lives in the browser. It has a Quake-style console (activated by opt+space) for quick chat and can analyze the content of any page you're on (e.g., YouTube transcripts, articles, other tweets) to help you create relevant content.
Engagement Predictor: A feature that scores tweet drafts in real-time to predict their potential for engagement. It's built on a model trained on my own dataset pulled from the X API and other public dataset from Kaggle[0].
Scheduled AI Calls: The system can call you on a predefined schedule to proactively brainstorm content ideas.
Here is the tech stack:
- Frontend: React, Tailwind, shadcn/ui
- Auth: X OAuth
- Payments: Stripe Subscriptions
- Voice AI: ElevenLabs Conversational AI, Twilio
- Engagement Predictor ML: Python, scikit-learn, XGBoost on a data pipeline from X API v2 and a base dataset from Kaggle.
- Chrome Extension: Same as Frontend and Chrome Extensions API
- Blog: Jekyll
- Infrastructure: Deployed on AWS Fargate using AWS Copilot for orchestration (ECS).
I'm building solo and just got the first trial user after 87 days of building in public. It's a long road but the feedback so far is encouraging.
Reading > listening for consumption.
Talk to create, read to consume.