Show HN: VoooAI – natural language to multi-modal AI pipelines
1 points
1 hour ago
| 1 comment
| voooai.com
| HN
Hey HN,Built VoooAI over 4 months. It's a NL2Workflow engine – you describe what you want in plain English, it generates an executable multi-modal pipeline (image + video + audio).The problem: Midjourney for images, Runway/Pika for videos, Suno for music – each has its own API, interface, and learning curve ComfyUI gives you control but requires understanding node graphs n8n connects APIs but doesn't understand creative intent What we did differently:Instead of having LLMs output JSON directly (hard to validate), we introduced a pseudocode IR layer: LLM generates pseudocode → deterministic interpreter expands to executable workflow JSON. This gives us: Syntax validation before interpretation Automatic engine correction (if skill requires specific engine) Reliable edge connections Try it: voooai.com Compare us: voooai.com/workflow-comparison – we're not better at everything, but for "idea → multi-modal output" in one prompt, that's our niche.AMA about the architecture.
▲
hkljjkl
1 hour ago
[-]
What Vibe Coding Taught Me Code review skills matter more than coding skills: I spend 70% of my time reviewing generated code, understanding it, catching subtle bugs. The AI writes fast but confidently wrong. Architecture conversations are gold: The most valuable prompts aren't "write this function" but "let's discuss the tradeoffs between X and Y architecture." Technical debt accumulates differently: I have less "I wrote this at 2am" debt, more "the AI suggested this pattern and I didn't push back" debt. Debugging is harder: When you didn't write the code, you don't have muscle memory for where things might break. Current State Not open source: The system integrates with 30+ APIs requiring paid subscriptions. Maintaining compatibility is a full-time job I can't delegate to the community. Business model: Freemium. 50 free points/month (10-20 images or 2-3 videos). Subscriptions for power users. What's working: NL2Workflow accuracy (~85%), multi-modal pipelines (image→video→audio), streaming UX What's not: Complex requests (>50 nodes), specific style transfer edge cases, real-time collaboration The One-Person Team Reality 4 months, one person, vibe coding. But let me be honest about what that means: I work 10-12 hour days The AI doesn't replace a team – it amplifies one person I still need to understand every subsystem deeply Customer support, marketing, DevOps – I do those manually Vibe coding let me compete on product scope, but it didn't eliminate the other hats a founder wears.
reply