# How we work

> We measure first, build in your environment, prove the result on real cases and then hand the system to your team.

- Page: https://www.ninalabs.ai/approach/
- Publisher: Nina Labs AI LLC

## Four phases

### 01. Map (1–2 weeks)

We sit with the people who do the work, measure the current process and pick the tasks where an AI system can finish most cases end to end.

Outputs: Ranked use cases; Baseline metrics; Architecture and risk notes.

### 02. Build (4–8 weeks)

We build the system in your environment: the agent or workflow, the tools, the permissions and the eval suite. Your engineers pair with us from day one.

Outputs: Working system in your cloud; Eval suite from real cases; Tracing and dashboards.

### 03. Prove (2–4 weeks)

The system runs in shadow mode on live cases while people do the real work. We compare the results and fix what the evals and the traces show.

Outputs: Shadow-mode results; Go-live criteria; Runbook and on-call guide.

### 04. Run (Ongoing)

We turn it on in stages, watch quality and cost, and re-run the evals on every model or prompt change. Then we hand it over to your team.

Outputs: Staged rollout; Monthly quality and cost report; Handover and exit plan.

## Engagement models

### Assessment (1–2 weeks · fixed scope)

We map one area of your business, measure it and tell you where an AI system would pay off, and where it would not.

- Process and system review
- Ranked use cases with business cases
- Architecture and risk notes
- A plan for the first build

### Build (6–12 weeks)

We build one system and take it through shadow mode to a staged go-live, in your environment and with your engineers.

- The agent, workflow or pipeline
- Eval suite and tracing
- Runbook and on-call guide
- Training for the team that will own it

### Embedded team (Quarterly)

Our engineers work inside your team on a shared roadmap, usually after a first build has shown results.

- A roadmap agreed each quarter
- Pairing with your engineers
- Shared standards and code review
- Monthly results review

### Run and improve (Monthly)

We watch the systems in production, re-run the evals on every model or vendor change and keep cost and quality on target.

- Monitoring and incident response
- Eval runs on every change
- Cost and latency tuning
- Monthly quality and cost report

## Principles

- **You own everything:** Code, prompts, eval sets, infrastructure definitions and documentation live in your repositories and your cloud accounts from the first day.
- **Measure before and after:** Every project starts with a baseline and ends with the same measurement. If the numbers do not move, we say so.
- **People approve what matters:** Irreversible actions go through a person until the evals and the production data show the system can do them alone.
- **No lock-in:** We choose models and platforms per task and build on open protocols such as MCP and A2A, so you can switch later.
- **Every engagement has an exit plan:** From week one we write down what your team needs to run the system without us, then we work through that list.
- **Small teams, direct contact:** You work with the engineers who build your system, not with a layer of account managers.

## What every system we ship includes

- **Evals:** A suite of real cases runs on every change to prompts, tools or models.
- **Tracing:** Every step, model call and tool call is recorded in OpenTelemetry format.
- **Approvals:** Actions that move money, change records or contact customers need a person until the data says otherwise.
- **Permissions:** Each tool has a written scope. Agents act with the user's rights or a narrow service identity.
- **Budgets:** Cost, token and latency limits per run, with alerts before they are reached.
- **Hosting:** Your cloud account or VPC by default. Open-weight models when data must stay on your hardware.
- **Models:** Chosen per task on measured quality, latency and cost. Switching is a configuration change.
- **Ownership:** Code, prompts, eval sets and infrastructure live in your repositories from the first day.
- **Handover:** A runbook, training for your team and a written exit plan in every engagement.

## Security and data

- We work in your environment, under your access policies.
- Client data is never used to train models. We use enterprise model agreements that exclude training on your data.
- Production data stays in your accounts. We use anonymized samples for evals where possible.
- We sign NDAs and data processing agreements before we see any data.
- Access is removed at the end of the engagement and confirmed in writing.

## Contact

Talk to an engineer through the [contact form](https://www.ninalabs.ai/contact/) or by email at contact@ninalabs.ai.