Engineer behind an endpoint accuracy fix in open-source Pipecat (59%→94%) and founding work at YC-backed Wayline joins Antler SF to build Miranda AI.
SAN FRANCISCO, CA, UNITED STATES, May 20, 2026 /EINPresswire.com/ — Kollaikal Rupesh, a voice AI engineer whose contributions to open-source streaming voice infrastructure include an endpoint detection fix that improved telephony accuracy from 59 to 94 percent, has been selected for the 2026 San Francisco residency of Antler, the global early-stage venture capital firm. Kollaikal joins Antler SF following his role as Founding AI Engineer at Wayline (Y Combinator S25), where he reduced conversational latency from multiple seconds to under 600 milliseconds and improved reliability in production, and the 2nd place finish of his startup Miranda AI at the Stanford CodeX LLM x Law Hackathon #6.
Kollaikal has contributed to Pipecat, the leading open-source framework for real-time voice agents, with over twelve thousand GitHub stars and production deployments at companies across the streaming AI ecosystem. His most significant contribution is a high-fidelity audio resampling change that improved endpoint detection accuracy on telephony audio from 59 to 94 percent — a measurable reliability gain across every Pipecat-based voice agent handling phone calls. Additional accepted pull requests include runtime failover behavior, heartbeat stability, and serializer reliability fixes. He is also the author of two open-source voice AI evaluation tools, Pipewatch and Coldcall, used to benchmark latency and transcript reliability of streaming voice pipelines.
“Voice AI has crossed the demo threshold and not yet crossed the trust threshold,” said Kollaikal. “The gap between them is closed by measurement and reliability work, not by bigger models.”
As Founding AI Engineer at Wayline (Y Combinator S25), Kollaikal led the streaming speech inference work that reduced end-to-end conversational latency from 4–5 seconds to under 600 milliseconds, crossing the threshold at which voice AI feels responsive rather than delayed. He also designed and built Wayline’s speech model evaluation framework, measuring word error rate stability, transcript drift, and latency distributions at P50 and P95 — the production reliability primitives most voice AI deployments lack. Since joining in early 2026, Kollaikal has become the top contributor to the Wayline codebase. Earlier, as a Forward Deployed Engineer at Smallest AI, he led production voice agent deployments to enterprise customers across multiple industries.
Kollaikal’s published research feeds directly into his engineering practice. His sole-authored survey, “Turn Detection in Production Voice AI Agents: A Survey of Approaches and Open Challenges,” taxonomizes production approaches across dominant open-source frameworks and identifies five open research challenges. He is also the author of “Toward Deployment-First Voice Clone Detection: A Lightweight and Robust Evaluation Protocol for Audio Deepfake Defense” (SSRN Preprint 6778759, 2026), and peer-reviewed work in the AIP Conference Proceedings (DOI: 10.1063/5.0233047).
Kollaikal has presented at AWS San Francisco on real-time healthcare AI agent design, co-hosted AI Dev Tools Night at Cloudflare HQ with Sourcegraph (251 attendees), and chaired Aggie Hacks 2025 at UC Davis.
In April 2026, Kollaikal founded Miranda AI, Inc., applying his voice AI infrastructure work to legal practice. Built in six hours for the Stanford CodeX LLM x Law Hackathon #6, Miranda placed 2nd among more than 650 attendees and was judged by representatives from Pear VC, Baker McKenzie, and Stanford Law faculty. Miranda has since expanded into a production platform that answers calls, runs legal intake with statute-of-limitations math and conflict screening, drafts engagement and demand letters live, and delivers qualified matters to attorneys. The platform is currently being deployed across plaintiff, litigation, and immigration practices.
About Miranda AI
Miranda AI is the AI operating system for plaintiff, litigation, and immigration law firms. One phone line answers every call, runs intake, drafts documents live, and delivers qualified matters to attorneys. https://miranda.legal
Kollaikal Rupesh
Miranda AI, Inc
rupesh@miranda.legal
Visit us on social media:
LinkedIn
X
Other
Legal Disclaimer:
EIN Presswire provides this news content “as is” without warranty of any kind. We do not accept any responsibility or liability
for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this
article. If you have any complaints or copyright issues related to this article, kindly contact the author above.
Media gallery
