Skip to content
Auzi

How we work.

Engineers who show up where the work happens, draw it as it really runs, and route every job to the right tool before building anything.

Five stations, every time.

Every engagement runs the same way, whatever its size. You keep something from each station — whether or not you keep us.

  1. 01

    Embed

    We start where the work happens.

    On site or inside your tools, sitting with the people who do the job. We learn the edge cases before we touch anything.

    Field notes

  2. 02

    Map

    Every workflow, drawn as it actually runs.

    Inputs, handoffs, waits and rework, drawn with your team — not the version in the handbook.

    Workflow map

  3. 03

    Route

    Each job gets a verdict and a reason.

    Rule, model, software or person, written down with the reasoning, the risk and the order to build in.

    Routing memo

  4. 04

    Build

    Small pieces, in use early.

    Working software in your team's hands early, not a big reveal. Anything a model does is measured before it ships.

    Working software

  5. 05

    Hand over

    You own it. We stay reachable.

    Code in your repository, infrastructure in your accounts, runbooks your team can follow. Stay on with us to run and improve it, or don't.

    Runbook

When to use AI, and when not to.

The calls we make most often. None of them is a law — each is where we start, and the map of your operation decides.

FIELD GUIDE — WHEN TO USE AIAUZI
The decision fits on an index card.Write the ruleRULEIf a person can explain it in two sentences, code can do it exactly, every time, for free.
The input is messy: scans, emails, photos, free text.Use a modelMODELReading and sorting unstructured input is where models earn their place.
It's the same judgement across a very large pile of similar cases.Model, with reviewMODELConsistent first drafts at volume, sampled and checked by the people who own the outcome.
A wrong answer is costly and hard to catch.Keep a personPERSONThe decision stays human. A model can gather the facts and draft the options.
It happens a handful of times a month.A checklistPERSONAutomation pays back on volume. A clear checklist and a capable person win here.
The data lives in several places and they disagree.Fix the system firstSOFTWAREA model reasoning over conflicting records gives confident wrong answers. Settle the source of truth.
The tool you already rent mostly fits.Integrate itSOFTWAREConnect what works and build only what it can't do. Rebuilding for its own sake is waste.
Nobody can describe how it's done today.Map it firstWe don't automate a process no one can explain. We sit with the people who do it and draw it.

Ways to work with us.

Start with an assessment, or bring a defined build. Every phase is written down before it starts: scope, deliverables and how we'll both know it's done.

  1. 01

    Assessment

    A written map and plan.

    A fixed-scope look at your operation that ends in a routing memo and a build plan you can take anywhere.

  2. 02

    Build

    A defined system, delivered in phases.

    Scope, deliverables and acceptance criteria written down for each phase before it starts.

  3. 03

    Embedded team

    Engineers inside your operation.

    A forward-deployed team working your backlog alongside your people, for as long as it's useful.

  4. 04

    Run

    After launch.

    Monitoring, model evaluations, fixes and improvements for the systems we built — or ones we inherit.

What's yours at the end.

Source code
In your repository, from the first commit.
Infrastructure
In your cloud and vendor accounts, not ours.
Data
In your database. Exportable, documented, yours.
Models & prompts
Versioned, with the evaluations that test them.
Runbooks
How it works, how to fix it, who to call.

How we handle your systems.

Least access
We ask for the access a task needs, in your accounts, and give it back when we're done.
Your requirements
We work under your NDA, security policies and review process.
Audit trails
What changed, who changed it and why — for code, configuration and automated decisions.
People in the loop
Any model that acts on your behalf has limits, logs and a human escalation path.

Bring us the messiest workflow you have.

An hour with the people who run it is usually enough to see where a rule, a model, software or a person belongs.

Book a working sessionhello@auzi.ai