Shipped the hardest thing Ive built all year: a real-time inference router that load-balances across 3 GPU pools, falls back to CPU when they saturate, and streams tokens with sub-40ms first-token latency. Spent two weeks just on the backpressure logic so a slow client cant stall the whole queue. The trick was a per-connection token bucket plus a priority lane for short prompts. Demo + write-up dropping this week. Ask me anything about the architecture.
Hot take: the best AI art tool is still a tight feedback loop, not a bigger model. I iterate 40 times before I keep one.
129
Generated an entire tarot deck with a fine-tuned SDXL LoRA — 78 cards, one consistent art style. The Tower came out terrifying.
98
Launched Dreambox — type a mood, get a generative ambient scene + matching soundscape. Built in a weekend.
94Prompt as Art entry: I wrote a prompt that describes itself being painted — recursive ekphrasis. The output is the prompt, illustrated. Recipe in thread.
7