My latest Blog Posts

My unpatched Next.js instances got hit by the React2Shell exploit within hours. Here's a complete breakdown of what happened, what the attackers tried and succeeded at, and why you must patch and rotate secrets immediately.

Learn how to set up a low-latency, streaming Text-to-Speech system using an Orpheus-style model, vLLM for efficient inference, and SNAC for audio decoding. This setup can run on a single 24GB GPU, or distribute the Orpheus model (vLLM/SGLang) and decoder (FastAPI/SNAC) across multiple GPUs.

Using GRPO to improve the TTS quality of a Llama based approach using Audio-Tokens with X-Codec 2.0. Implementing a custom reward using Word Error Rate with Whisper and a custom repetition penalty.

Azure removed Bing API resources, forcing users to switch to a restricted search tool within the AI Agents Service under stricter terms and data policies.

Comparison of different quantization techniques and their impact on inference time. And how to optimize inference time with quantizations using vllm.