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
End-to-end GitHub Actions workflow for an OpenAI Realtime + LiveKit voice agent: matrix build, eval gate, cosign + SLSA provenance, and a kubectl rollout to k3s. Real YAML included.
Open Dental and Dentrix run 60% of independent US practices but ship no native chat. Here is how 2026 chat agents bolt onto their APIs to cut no-shows by 30% and recapture missed-call revenue at $290 per appointment.
One routing brain, many narrow workers. Why supervisor-and-specialists wins in production over decentralized swarms — with LangGraph code, OneRoof's 10-agent topology, and the failure modes nobody warns you about.
A working pgvector quickstart for AI agents — install the extension, design a hybrid table, build an HNSW index, and query top-k matches from Prisma and Python. Real schema, real code, no marketing fluff.
Conversational forms convert 40% better than static ones. Here is how 2026 chat agents inline name, email, qualifying, and consent fields into the thread without breaking flow.
Natural conversation turn-taking lands at 200-300ms. We break down the full sub-500ms budget across VAD, ASR, LLM TTFT, TTS, and network — and show where each ms goes in CallSphere's Healthcare Realtime stack.
Bidirectional Media Streams ship raw mulaw both directions over a single WebSocket. We break down the four patterns CallSphere ships in production: proxy-to-OpenAI, sidecar STT, conference fork, and replay-on-reconnect.
Default WebRTC ICE gathering still leaks private LAN IPs to remote peers. mDNS obfuscation, host candidate filtering, and STUN/TURN forced relay close the gap for production AI voice apps.
Tool-calling agents drift on edge cases your prompt cannot fix. We walk through the OpenAI SFT recipe for gpt-4o + gpt-4o-mini in 2026, the JSONL format with `tools` arrays, strict-mode caveats, and a CallSphere-tested checklist for hitting 95% function-arg accuracy.
