Eddie Siegel

Eddie Siegel is chief technology officer of Ode with Anthropic and co-founder of Fractional AI, the applied-AI engineering company acquired to establish Ode. He builds enterprise systems that turn powerful but unpredictable models into dependable software for consequential business operations.

From data infrastructure to enterprise AI

Siegel studied computer science at the University of Pennsylvania and worked at LiveRamp and in engineering leadership roles connected to Rapleaf and TowerData. He co-founded the marketing-data integration company Wove, which Samba TV acquired; at Samba TV, he became vice president of product and engineering. He later co-founded Xip with Chris Taylor and served as CTO, building tools for creators to work asynchronously with customers.

Siegel and Taylor subsequently launched Fractional AI to build customized enterprise AI systems, combining model capabilities with operational expertise, production engineering, and integration into existing business processes. His work with Airbyte illustrates the practical demands of AI-generated API integrations: inconsistent documentation, hallucinations, caching, security, testing, and model costs.

Ode acquired Fractional AI in May 2026 and launched that July, with Taylor as CEO and Siegel as CTO. Backed by Anthropic, Blackstone, Hellman & Friedman, and other investors, the company applies frontier models and engineering teams to enterprise-wide operational challenges.

Building voice agents that hold up in production

Siegel’s production voice-agent case study describes an AI interviewer for qualitative research inside large organizations. Its architecture replaces a single sprawling prompt with interview goals, deterministic question management, tool calls, and background text agents.

• Conversational freedom needs explicit controls. A drift detector tracks whether a question has been answered or the interview has wandered; a next-question agent selects follow-ups using the conversation’s broader objectives. Together, they preserve improvisation while keeping research interviews productive.

• Transcription reliability is a product requirement. Because an audio model and its separate transcription system can interpret the same silence or background noise differently, contextual checks identify suspicious transcript segments and hide misleading text from the interface.

• Synthetic-conversation evaluations make subjective quality measurable. Simulated interviewees with different personalities and job functions generate test conversations scored for clarity, completeness, and professionalism. Although these assessments lack definitive ground truth, they expose edge cases and prompt regressions before changes reach users.