So, I built Ask-AI.
How it works (The Tech) Instead of a simple wrapper, I built a routing engine in Node.js/Vercel that analyzes user intent before sending the request: News/Current Events? → Regex detects "news", "price", "latest". Routes to Grok 3. Video Prompt? → Detects "create video", "animate". Routes to Luma Dream Machine. Audio/PDF Upload? → Routes to Gemini Pro (for its native multimodal context window). Logic/Code? → Routes to GPT-4o, GPT 5.2 or o1.
The Launch Stats (Day 10) I soft-launched 10 days ago on the Play Store. Installs: 1,170+ (100% organic/ASO). Conversion Rate: 24.9% (Store listing visitors -> Installs). CAC: $0.00.
The "Retention Bug" My retention looked terrible at first (~20% user loss). I dug into the logs and realized ~170 uninstalls came specifically from Russia and Iran, where the OpenAI/Gemini APIs are geo-blocked. My app was crashing for them. I’ve since geofenced the app, and retention in supported regions has stabilized around ~30%.
Features I'm testing: Podcast Mode: Takes any chat thread/document and generates a 2-way audio discussion (like NotebookLM but integrated into the chat). Voice Journal: Hands-free voice notes that auto-tag and save to a vector store (Knowledge Base). I’m looking for feedback on the routing logic—does it feel seamless, or can you tell when it switches engines?