Systems we built and ran ourselves.
Seven working systems, debugged at our own expense before earning a page here. Engagements start from this infrastructure, not a blank repository, and land in your cloud with our engineers beside your team.
Multi-agent research with a live audit trail: a human spend gate before dispatch, source verification before writing, fabrication review before delivery, and a graph you can watch work.
A one-line prompt becomes a self-healing browser test suite: cross-model assertion consensus, selector caching, autonomous monitoring between releases. Healing repairs the path, never the assertion.
Ask a video library a plain-language question and get the exact clip back. Multimodal embeddings fuse what is seen, heard, and said, searched at scene level by specialist agents.
An autonomous pipeline from territory to booked meeting: scan, diagnose every business with AI, score the fit, run grounded outreach. Models only where judgment is needed, SQL everywhere else.
Customer-facing agents with adult supervision: a 24/7 support agent and a marketing rep streaming in real time, plus a command center showing the business every conversation held in its name.
A team knowledge workspace: chat, structured memory, notes, bookmarks, and a document inbox. Retrieval spans all of it, so “what did we decide about X?” gets an answer.
Safety-critical AI engineering in a kitchen: recall matching that can never say “safe,” dietary filters enforced in code, not prompts, and vision that asks before it writes.
Everything here began as a lab project run at our own expense. Eleven projects are on the bench now, including STG — the successor to AutoTest.