Built core AI agent and LLM features for a generative image and video platform, with a strong focus on multi-provider orchestration, prompt optimization, and Cloudflare-based infrastructure.
- Integrated multiple model providers (Google Gemini/Vertex, DeepSeek, Kling) with per-provider token caps, JSON-mode enforcement, and graceful fallbacks.
- Optimized the system prompt for cost and latency with per-turn section gating and prompt-cache telemetry, and made gallery queries sargable via functional Postgres indexes.
- Resolved Cloudflare Worker scaling issues by moving heavy autorun work to Trigger.dev, eliminating out-of-memory 503s, and shipped a hardened analytics proxy worker with rate limiting.







