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
How the Postgres MCP server exposes schemas and queries to LLMs without letting them rewrite your tables. Read-only transactions, scoped roles, and the patterns CallSphere uses across 115+ tables.
Zero-cloud voice agent on a single laptop: whisper.cpp for STT, llama.cpp for the LLM, and Piper for TTS. No telemetry, no API keys, no per-minute bill — full working code.
ClickHouse 26.2 added time-based block flushing and ClickPipes covers Kafka, S3, and Postgres CDC natively. Here's how CallSphere streams 50k voice agent transcripts a day into a ClickHouse cluster with sub-second p95 query latency.
Porting an existing business line to a Twilio-backed AI receptionist takes 2 to 8 weeks, an LOA, and a CSR pull. Here is exactly what to expect, what blocks ports, and how to keep zero downtime during cutover.
Golden sets are the bedrock of every serious voice agent eval. Here is how we mine real call traffic, label intents, freeze edge cases, and version the dataset that gates every CallSphere release.
Sizing questions drive 50 to 70 percent of fashion returns and cost $2.70 to $5.60 per ticket. Brands deploying AI fit advisors see 33% return drops and 25% AOV lifts. Here is how chat agents fix apparel D2C in 2026.
Build a voice agent that runs Whisper, Deepgram Flux, Aura-1 TTS, and Llama 3.3 on Cloudflare's 330-city edge network. Architecture, code, latency budget, and CallSphere-grade pitfalls.
POLQA (ITU-T P.863) replaced PESQ for wideband and super-wideband AI voice. Here is how we score MOS in production, why 4.0+ is table stakes, and how to wire perceptual quality into a Twilio call pipeline without a phone lab.
BM25 alone hits 65% recall@10. Dense alone hits 78%. The hybrid pipeline pushes 91% — and once you bolt on a reranker the gap widens. Here is how CallSphere wires hybrid retrieval into a voice loop with a 200ms budget.
