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
Stop running brittle nightly batches. Postgres logical replication gives you sub-second CDC into a warehouse, feature store, or vector index. A working publication, slot management, and a Supabase ETL Rust example.
GPT-4o native image gen, Flux 2, and Stable Diffusion 4 power conversational image creation. Here is how 2026 chat agents render images inline, iterate on prompts, and edit in place.
ElevenLabs Flash v2.5 hits 75ms inference; Cartesia Sonic streams under 100ms. We tune chunk_length_schedule, sentence-boundary streaming, and TTFB to keep TTS under 200ms of the voice budget.
Chrome M116+ joined Firefox and Safari with one-time mic/camera permissions in 2026. For voice AI, that means redesigning your prompt UX or losing 30%+ of first-time users mid-call.
DPO replaces the entire RLHF pipeline with a one-line classification loss — no reward model, no PPO, no instability. Both OpenAI and Azure now expose DPO endpoints. Here's how to layer it on top of SFT for tool-using agents.
Agentic refund chat reaches 70–92% deflection but trust craters when users get stuck in maze loops. Here is the 2026 playbook for refund and RMA chat that resolves, not deflects.
"You are an expert" prompts dropped MMLU accuracy from 71.6% to 66.3% in 2026 benchmarks. We translate the research — when persona helps (alignment, voice, refusal), when neutral wins (math, code, classification) — into the routing rule CallSphere uses across 37 agents.
Analysis of how Accenture and Databricks help enterprises deploy AI agents using data lakehouse architecture, MLOps pipelines, and production-grade agent frameworks.
How agents communicate in multi-agent systems using direct message passing, shared blackboard, event-driven pub/sub, and MCP-based tool sharing with production code examples.
