Todd Fisher

Todd Fisher is head of software engineering at Philo Ventures and the former founding chief technology officer of Verdi AI. His career spans healthcare software, engineering leadership, startup formation, generative-AI reliability, and a musical experiment that turns an electric guitar into a speaking, conversational instrument.

From healthcare software to AI startups

Fisher studied at Southern Utah University and joined TouchMD as a founding engineer, eventually becoming director of engineering while developing patient-education software. At Pluralsight, he served as principal of software engineering practices in its Technology Center of Excellence, helping engineering teams improve collaboration and product development.

As Verdi AI’s founding CTO, Fisher confronted testing nondeterministic AI applications: model responses can vary unpredictably, exposing failure modes that conventional software tests miss. His work on generative-AI reliability examined practical ways to test ChatGPT-powered features and protect surrounding applications from unstable outputs.

At Philo Ventures, Fisher helps founders and engineering teams build early-stage products. He has also helped organize a Utah community for startup-oriented engineers and technical leaders.

Making an electric guitar speak

Fisher’s speaking-guitar project grew out of a Halloween application that translated guitar notes into illuminated letters. He subsequently built a JUCE plug-in for Logic Pro that combines speech synthesis, pitch detection, language models, and processed vocal recordings.

• A guitar as a conversational interface: Microphone input passes through Whisper speech recognition and a local language model; the guitarist triggers fragments of the generated response by playing notes.

• Editable speech segmentation: Natural speech does not reliably separate words with silence. After experimenting with energy gaps and syllable-related peaks, Fisher added manual controls instead of assuming automatic segmentation would work consistently.

• YIN pitch-detection algorithm: Fisher identifies each played note’s fundamental frequency, creates a corresponding sawtooth-based synthesized tone, and shapes it with ADSR controls before combining it with speech.

• Precomputed vocal processing: Using VocalSet recordings and the WORLD audio-processing project, Fisher pitch-shifted sung samples and mapped them to guitar notes. Because the transformations were too computationally expensive for live performance, he prepared them in advance.

The project treats AI as material for hands-on experimentation: its value lies in building a playable instrument while confronting latency, imperfect automation, and the practical limits of real-time audio.