Learn to build voice agents.

A Gemma voice-serving case study: 1,520 measured requests, concurrency trade-offs, connection reuse and a downloadable analysis kit.

Understand NVFP4 with Gemma 4 26B A4B: weight memory, hardware kernels, FP8 KV cache and the experiment needed to prove a latency improvement.

Use our archived concurrency sweep to understand vLLM batching, tail latency, prefill budgets, warmup and goodput for voice workloads.

Measured shared-prefix results, KV capacity arithmetic and a practical context budget for voice agents using Gemma and vLLM.

Which Frequency API to start with: phone agents, the Realtime API, Fourier TTS or transcription. What each one does, and how a call works.

A hosted voice-agent API, a speech-to-speech Realtime API, or a stack you build from ASR, LLM and TTS: control, latency, cost and who runs it.

How we bill voice agents, Realtime sessions, TTS and transcription: rates, 10-second blocks, what's free, commitments, and worked monthly estimates.

Place a real AI phone call in Hindi, Tamil or Telugu from Python, then read the transcript off a signed webhook. Three requests, with httpx and Flask.

Upload a CSV this afternoon and an AI agent calls every row in Hindi, Tamil, Telugu and more, inside your hours, with retries and a result per contact.

Hear about every call the moment it changes. Verify the signature on the raw body, store before you answer, ignore retried duplicates, order by status.

Split the delay a caller hears into turn detection, the model, TTS first byte, the network and playback, and read what Mira's request log reports.