Drop Aksel into Claude and get a senior voice AI engineer who builds low-latency realtime agents where latency is the product.
Aksel builds voice and realtime agents: the STT to LLM to TTS pipeline (and speech-to-speech realtime APIs), the latency budget (sub-800ms target broken down across VAD, endpointing, STT, LLM TTFT, TTS first-audio and network), streaming everything, turn-taking (voice activity detection, endpointing, barge-in and interruption), telephony vs web (SIP, WebRTC, codecs), mid-conversation tool calls, error recovery, and evaluation (latency percentiles, word error rate, task success). Tools like Deepgram, Whisper, ElevenLabs, Cartesia, LiveKit, Pipecat and the OpenAI Realtime API, without lock-in. Test in staging under real network conditions before production.
What you get
-
→The realtime STT-LLM-TTS pipeline and a per-stage latency budget
-
→Turn-taking: VAD, endpointing and barge-in handling
-
→Telephony vs web (SIP, WebRTC) and mid-call tool calls
-
→Evaluation: latency percentiles, word error rate, task success
📄 aksel-voice-realtime-agent-engineer.skill
Under 2 min install
Works with Claude, ChatGPT & any AI chat
How to install
Download the .skill package → open Claude → paste SKILL.md into your Project Instructions or system prompt → describe your requirement → Aksel builds the answer. Includes a full worked example so you see exactly what you get.