Hakob Astabatsyan, Co-founder & CEO, SynthFlow AI | Slush 2025
summary
Claims a practical path to production for voice AI using the build, evaluate, launch, learn framework: a visual conversational flow builder that pairs flow logic with LLM prompting, large-scale simulations on synthetic data to run thousands of test calls, a custom AI-native telephony stack that reduces latency and improves reliability, and feedback loops to fine-tune models. Reports processing millions of calls, 24% answered outbound calls and 15% closed deals, with unit economics beating human SDR costs.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:24Voice AI last mile problem
- 3:37Bell framework overview
- 3:59Build step visual flow builder
- 5:22Evaluate step simulations
- 6:26Launch step AI telephony system
- 7:08Learn step data and KPIs
- 8:33Innovation beyond demos
- 9:15Culture building in AI companies
related talks
01h 02m[Full Workshop] Building Conversational AI Agents
01mGovAI Summit Spotlight!
44mSmartly AI: Cross Platform Voice Application Creation
56mAI Agents That Don’t Suck: How to Build Ones That Actually Drive Business Value
17mBuilding Effective Voice Agents
19mRise of the AI Architect
51mBuilding an AI assistant that makes phone calls
19mThe Advantages of Building Your AI Stack In house