Case study · Project Operations
NeivuX
AI-Run Project Platform for Small Teams
The brief
NeivuX is a project platform for small teams that are not software teams — the agency, the trades company, the consultancy, the clinic office — built on the premise that the method should fit the work rather than the other way round. A project is created under one of twelve method templates (waterfall, Scrum, Kanban, Scrumban, hybrid, stage gates, critical path, Lean DMAIC, Shape Up, OKR, recurring operations, personal) and is then readable as a list, a board, a table, a calendar, a Gantt chart or a workload view, each a real deep link. Progress and the Gantt schedule are computed rather than typed, including the critical path, and WIP limits warn instead of refusing. Underneath runs a server-side loop: eight automated watchers flag what is late, stuck, aging, over its WIP limit, silent, unassigned, slipping or on the critical path; a scheduled daily report goes out each morning; and recurring operations spawn on their own cycle. The AI layer is deliberately held back from the database — the planner proposes a plan of at most twenty-five tasks, a DAG by construction, and a person presses Approve before anything is written, under the same membership rules a human write obeys. Beside it sit a streaming assistant on two Vertex Gemini tiers (instant and deep reasoning), long-form verbatim voice dictation that segments at pauses and runs to sixty minutes, a policy-driven case assistant that drafts replies with optional PII redaction and PDF export, and a Drive-synced knowledge base with Vertex embeddings, hybrid retrieval, citations and abstention.
The problem
Project tools are built for software teams. An agency, a trades company or a clinic office gets handed Scrum semantics it does not run, a board that does not know what is late, and an AI bolted on the side that can only chat — so the plan lives in somebody’s head and the status meeting is how anyone finds out.
Our approach
We built the method in rather than on top: twelve templates, six views over the same project, a schedule and a critical path the product computes instead of asking a person to maintain, and a server-side loop of watchers that reports what is late, stuck, aging or unowned without being asked. The AI plans and drafts, but the write path is gated — the planner proposes at most twenty-five tasks as a DAG and a person approves before anything lands, under the same rules a human write obeys.
The outcome
A project is created under the method that fits it and is then legible six ways, with progress and the critical path computed rather than typed. The watchers and the daily report run on a schedule in production, and the AI never writes to the board on its own — so what the plan says is what a person agreed to.
What we built
Tech stack
Key metrics
Want one like this?
We can design, staff, and build a platform like this for your operation.
Get started