Independent AI builder in the Chicago area
I'm Drew Sepeczi. I build useful AI systems for real businesses.
Vault Zero is the studio I run to turn ambitious AI ideas into dependable products. The current business focus is AI receptionists for home-service teams that need missed-call, overflow, and after-hours intake handled with clear rules and human handoffs.
I also design, build, and operate my own software ventures. They are practical experiments in how people can create interfaces, coordinate agents, work with private knowledge, and keep control of AI systems close at hand.
The business
One studio, two complementary kinds of work.
- AI receptionists for home services. Voice systems for plumbing, HVAC, roofing, and electrical teams, designed around the calls and handoffs that matter to each business.
- Full-stack AI product work. Strategy, interface design, engineering, and production integration for founders and product teams with a focused problem to solve.
- Products operated firsthand. The ventures below keep the work grounded in real release decisions, customer experience, reliability, and iteration.
Selected ventures
Products I'm building across the AI stack.
Squid Agent
An AI product workspace for turning prompts, screenshots, and URLs into working React prototypes with evidence built in.
Nodebase
An open-source visual studio for building, running, and sharing AI workflows across agents, tools, and approval steps.
RagBase
A private document workspace for cited answers from PDFs, contracts, notes, and webpages without signup friction.
Trace
A local-first desktop workspace for running AI agents, managing files, using skills and MCP tools, and keeping approvals under control.
How I work
Ship something real. Be honest about what it proves.
I work across product strategy, interface design, engineering, AI orchestration, and the unglamorous integration details that decide whether a system holds up in production.
Voice systems get explicit safety boundaries, transfer rules, and approved business logic. Product work gets the same care around data, permissions, failure states, and human control.
I separate a working demo from a deployed product, and a deployed product from a measured customer outcome. That distinction keeps the work useful and the claims honest.
Work together