Instead of static articles or black-box optimizers, you pick a panel of up to six AI analysts (tuned on different strategies: GPP, data-heavy, contrarian, etc.). Ask a question like “Tua or Allen?” → they argue it out in real-time with ElevenLabs voices. The transcript shows why each pick wins, and you can export the chosen lineup directly to DraftKings/FanDuel.

Tech stack: • Backend: Cloud Run + Cloud Functions (serverless-first) • Frontend: Next.js (hosted via Lovable) • Database: Firestore (real-time debate/live dashboard) • Caching: Redis + Cloudflare edge routing (infra <$0.05 per debate) • Models: six LoRA adapters trained on open-source LLMs (Vertex AI preemptible GPUs) • TTS: ElevenLabs for character voices

Live here: https://fantasyfootballneuron.app

Would love feedback from the HN community on the approach — especially around scaling real-time debates and caching optimizations.