Vish Kulkarni

About

As part of the forward deployed team at ElevenLabs, I work with enterprise customers across industries to design, build, and launch voice agents that carry real workload. Support lines, scheduling, intake, outbound follow-up. Off-hours, I run my own AI research infrastructure to stay hands-on with the technology I evaluate. Personal AI infrastructure (built on personal time + equipment): → Autonomous agent orchestrator running on a Mac Mini coordinating ~30 specialized agents (research aggregators, content drafters, daily digests, monitoring loops). All sharing one MCP-based memory substrate - Postgres + pgvector + tiered summarization. Patterns informed by Letta sleep-time agents, AgeMem typed memory ops, and openhuman's memory tree, re-implemented with measured improvements over each. The goal: understand multi-agent coordination at the implementation level. → Voice cascade architecture for hold-to-talk interaction with frontier LLMs: whisper.cpp local STT → Anthropic API → ElevenLabs / kokoro-onnx TTS, accessible from iPhone via Tailscale HTTPS. Extended into a voice-native kid story platform that generates illustrated stories on the fly. Built to learn the latency, the failure

Experience

Education