A product laboratory for consumer voice AI: a personal assistant that turns natural-language requests into real phone tasks while only exposing sensitive data when the task actually requires it.
My role
Founder, Product Owner, AI Architect & Full-stack Engineer
Task history showing completed and failed real-world call outcomes across personal errands, bookings, and follow-up requests.
01
Product laboratory
Built from scratch as a consumer voice-AI product, spanning the application, AI behavior, voice execution, privacy model, and deployment.
AI Call Center proves the B2B operating system behind production voice agents. AI Caller explores the consumer edge: what changes when the same execution patterns are applied to personal errands, family coordination, clinics, restaurants, and daily follow-up.
The product is built around a simple promise: a user should be able to describe a task naturally, let the assistant decide what context is needed, place the call, and get back a usable outcome instead of a transcript-shaped dead end.
02
Consumer constraints
Delegation
The assistant has to turn vague everyday requests into concrete tasks that can be scheduled, executed, retried, or resolved later.
Trust
Personal data is more sensitive in a consumer product, so context cannot be loaded indiscriminately just because it might be useful.
03
Execution, not just planning
The dashboard keeps the product grounded in outcomes. Each task becomes a visible operational record with status, target, result, and failure reason, so the user can see what actually happened and what still needs attention.
That lifecycle matters in consumer voice AI because the real work often includes missed calls, rejections, no-availability states, and tasks that must be picked up again later.
04
Structured context
Contacts give the assistant grounded targets instead of forcing the user to restate names, numbers, categories, and notes every time.
That small layer of structure is what lets a consumer assistant feel useful in practice: a dentist, a tennis center, a restaurant, or a family contact becomes reusable operational context.
Contact book storing reusable people and businesses with categories, phone numbers, and task-relevant notes.
05
Privacy boundary
Household profiles separate ordinary context from sensitive information. The assistant can identify who a task is about, but vault-style data stays hidden by default and is only exposed when the task actually requires it.
That boundary is the core product thesis: a personal assistant should be able to act with real context without turning every task into blanket access to private data.
Household profile editing with personal context visible while sensitive vault data remains hidden by default.
06
Task path
The runtime path stays simple from the user's point of view while enforcing structure under the hood.