Sattaya.AI — Blog by Sattaya SingkulWriting on end-to-end AI products, voice assistants, LLM agents and on-premise AI in production.https://joessattes.github.io/Sattaya-AI-Website/enReading a model's architecture from its safetensors header, without downloading the weightshttps://joessattes.github.io/Sattaya-AI-Website/blog/reading-the-architecture-off-the-weights/https://joessattes.github.io/Sattaya-AI-Website/blog/reading-the-architecture-off-the-weights/How I recover a neural network's anatomy from tensor names and shapes alone: 96.8% role accuracy on 510 architectures, from about 46 KB of header. The method behind AI Model Explorer.Mon, 02 Nov 2026 00:00:00 GMTResearchOpen sourceQuantizationWhy emotion models fall apart outside the lab (and what actually helps)https://joessattes.github.io/Sattaya-AI-Website/blog/emotion-models-outside-the-lab/https://joessattes.github.io/Sattaya-AI-Website/blog/emotion-models-outside-the-lab/Speech emotion recognition looks solved on benchmarks and breaks on real calls. A researcher's tour of domain shift, speaker leakage, metric learning and self-supervised features.Tue, 13 Oct 2026 00:00:00 GMTResearchVoice AIFinOps-Estimator: sizing GPUs for a voice AI pipelinehttps://joessattes.github.io/Sattaya-AI-Website/blog/finops-estimator/https://joessattes.github.io/Sattaya-AI-Website/blog/finops-estimator/An open-source tool that turns concurrency and model choices for STT → LLM → TTS into VRAM, RAM, CPU, TFLOPS and cloud or on-prem cost.Tue, 13 Oct 2026 00:00:00 GMTOn-prem AILLM servingOpen sourceFitting an LLM in your own server room: the memory math behind on-prem AIhttps://joessattes.github.io/Sattaya-AI-Website/blog/on-prem-llm-memory-math/https://joessattes.github.io/Sattaya-AI-Website/blog/on-prem-llm-memory-math/Bits per weight, quantization error, outliers, KV cache and memory bandwidth: how to decide which model actually runs on the GPUs you own.Tue, 13 Oct 2026 00:00:00 GMTOn-prem AIQuantizationLLM servingTurboQuant, reviewed: when a theory paper walks straight into productionhttps://joessattes.github.io/Sattaya-AI-Website/blog/turboquant-review/https://joessattes.github.io/Sattaya-AI-Website/blog/turboquant-review/A close read of TurboQuant (ICLR 2026): random rotations, Lloyd–Max codebooks and a 1-bit QJL residual, and why it went from proof to vLLM and a 17k-star vector index within weeks.Tue, 13 Oct 2026 00:00:00 GMTResearchQuantizationLLM servingYour MAU is not your load: capacity math for voice AI at scalehttps://joessattes.github.io/Sattaya-AI-Website/blog/voice-ai-at-scale/https://joessattes.github.io/Sattaya-AI-Website/blog/voice-ai-at-scale/How to turn "a million users" into a latency budget, a concurrency number, KV-cache memory and a GPU count — with the formulas, not vibes.Tue, 13 Oct 2026 00:00:00 GMTLLM servingVoice AI