Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
Drachtio is connect-style middleware for SIP, with a C++ Sofia core and a Node.js framework on top. In 2026 it is the fastest path from carrier trunk to OpenAI Realtime if you live in JavaScript and rtpengine.
Google now hosts an official Calendar MCP. We compare it against nspady/google-calendar-mcp and show the booking loop CallSphere voice agents use to schedule appointments end-to-end.
faster-whisper handles STT on CPU; vLLM serves a 70B model on a single H100 with 6x throughput vs HF Transformers. Here's the production-grade voice pipeline that connects them.
An STT-LLM-routing pipeline that detects intent on every utterance and routes to the right specialist agent in under 500 ms. We cover schema, debouncing, and how CallSphere routes 6 verticals worth of intents.
Calendly schedules, Zoom hosts, WebRTC fans out, AI transcribes. Here is the 2026 reference architecture for interview transcription with speaker diarization, sentiment, and CRM hand-off.
SWE-bench measures whether code agents can resolve real GitHub issues. The methodology - human-verified, repo-grounded, deterministic - is exactly what voice eval needs to copy.
Sub-second voice agent with zero idle cost: API Gateway WebSocket ($connect/$disconnect/$default), Lambda per-message handler, DynamoDB session store, and OpenAI Realtime over WebSocket.
The online pet care market hits $36.4B in 2026 with subscription-driven loyalty. AI chat reduces pet D2C support tickets by 70% on subscription edits, allergens, and feeding guides. Here is the playbook.
Deploy Kokoro-82M TTS, Whisper-large-v3-turbo, and CosyVoice2 on Replicate's GPU edge with cog-based packaging. Cold-start and cost analysis for production voice agents.
