Skip to content

Senior Analytics Engineer

A company’s HR data lives in five or six systems that don’t talk to each other. Zivo solves this with the Data Hub: a ready-made, production-proven HR warehouse that connects to any combination of systems and answers both people and AI agents. We’re looking for a senior person to own the data model and the agents that build it.Zivo is a startup on a mission to make HR AI-native through AI journeys; we do it by productizing the operations of an HR team. The Data Hub is its People Analytics front: the single, trusted foundation on which people decisions are made. The buyer is the CHRO; the people who use it every day are HRBPs, compensation analysts, recruiting and people ops.HR has the data. It just can’t use it.Today, for the HR team. HRIS, payroll, recruiting, performance and spreadsheets, each with its own format, its own key and its own version of the truth. Half the questions that matter require joining three of them, by hand, over days. When it’s done, the number doesn’t match finance’s.

Product/Engineering · Remote

“How much does payroll cost by department, including payroll taxes?”

“What is voluntary turnover in the first six months, by manager?”

“Who is below the salary band for their role?”

With the Data Hub. The company connects its systems and, within weeks, has its own warehouse built on an HR model that has already solved the problems it has yet to discover. The same questions come out through a dashboard, SQL or an analyst agent, always with the same number. Bad data at the source becomes a report that says what, why and whose fix it is. Who sees what, from salary to performance review, is decided by HR.Off-the-shelf People Analytics tools leave the modeling to the customer; generic BI gives infinite flexibility and zero opinion. We bring the model and the people who adapt it to each customer.Extraction from source systems sits with the platform team. This role owns everything from Raw onward.

Model before configuration. We show up with a ready, production-proven model; every system integrated once serves every customer that follows. The customer models nothing: they connect.

Native Brazilian reality. CPF, matrícula, CLT and PJ, dissídio, 13th salary, payroll taxes: modeled, not translated. LGPD out of the box.

Trust is product. Status, quality findings and permissions are features the customer sees, not backstage work.

Specialists at the edge. Zivo people do the last mile with each customer. The model is their lever.

What you’ll do

1. A warehouse builder, not a warehouse

Every customer has its own industry, its own maturity and its own mix of systems, from an HRIS with an API to a spreadsheet maintained by hand. We don’t build one model per customer: we build the machine that, for any combination of sources, produces a warehouse with the same model, the same modules and the same consumption contract. The core stays generic, what’s specific to each customer stays isolated, and onboarding becomes a repeatable operation, not a project.

For the customer: off spreadsheets in weeks, on a model that has already solved what they have yet to discover.

Day to day: the same field arrives as a spreadsheet column at one customer and as an API custom field at another; the core serves both without getting contaminated.

2. The build is agentic

Integrating a system, auditing a customer, migrating from one system to another, evolving the model: each flow is a skill that an AI agent runs. Instead of doing the integration, you design and tune the process that does it: what the agent needs to know, which decisions it doesn’t make on its own, how it proves it got it right. Every incident becomes a rule, a test or a guardrail in the process, never a patch on a single customer.

For the customer: same quality and predictable timelines, no matter how many customers arrived the same week.

Day to day: an entire onboarding runs from a single command, with seven human decision points; each one exists because skipping it once caused an incident.

3. A model for people and, above all, for agents

Final tables and the metrics layer are consumed by dashboards and by an analyst agent that answers HR questions in natural language. An agent is a demanding consumer: unambiguous names, column descriptions that carry the meaning (the catalog is the agent’s interface), standardized values in Portuguese, explicit grains, one definition per metric. The model needs to be efficient to query and impossible to misread.

For the customer: anyone in HR asks and gets the right answer, with the same turnover definition that’s on the CHRO’s dashboard.

Day to day: an ambiguous column description isn’t a documentation detail; it’s the agent giving the CHRO the wrong answer.

4. Quality and reconciliation as a discipline

Systems overlap: the same person exists in the HRIS, in payroll, in the ATS and in a spreadsheet, with names, dates and employee IDs that drift apart over time. Reconciling identity (CPF, system id, matrícula), deciding which source wins for each field and from when, handling rehires and CLT to PJ conversions without inventing terminations, separating the data we lost from data that never existed at the source. Every rule becomes a test; every finding becomes an audit line with an owner and a severity, visible to the customer.

For the customer: numbers that match each other and match payroll; when they don’t, a report says why and whose fix it is.

Day to day: a customer switches payroll providers mid year; history comes from the old system, the present from the new one, and the seam closes person by person.

5. Granular permissions over sensitive data

Salary, CPF, termination, performance review: a CHRO, an HRBP, a manager and the employee themselves need different views of the same data, identified or anonymized, restricted to their part of the organization, by module, and LGPD requires that this be demonstrable. All of it at the column and row level, failing closed, without duplicating the model and without the agent handing over what the person isn’t allowed to see.

For the customer: HR decides who sees what, and the rule holds the same across dashboard, SQL or agent.

Day to day: a row policy that fails closed returns zero rows with no error; telling “no access” apart from “no data” is part of the design.

What will you do?You’ll own a model that dozens of companies use to make decisions about people, with AI agents in the daily engineering flow and direct contact with specialists and customer HR teams.

Data model: specs for each module, vault, canonical tables and metrics layer, from design to evolution with controlled impact across all customers.

Skills and agents: design and tune the onboarding, integration, audit and QA flows. Every incident becomes a rule, a test or a process guardrail.

New systems: integrate APIs, spreadsheets and exports into the shared model, keeping the core generic.

Quality and reconciliation: identity, source precedence, system switches with cutoff dates, history. Validate by walking through real records, not just aggregates.

Permissions: column governance, row policies, identified and anonymized variants, and how the agent respects all of it.

Desired profile

Seniority here is less about years and more about having already been the person who sorted out the inconsistencies in an environment with many systems and many entities.

Essential

Dimensional modeling.

Data Vault 2.0 modeling: hubs, links, satellites, bitemporality.

Pipelines with multiple entities and multiple sources.

Agentic flows: you’ve built or operated flows with LLM agents (skills, prompts, decision points, evaluation).

Semantic layers and dbt.

Pipelines with strong data quality requirements: tests, reconciliation, auditing.

We take as a given

Advanced SQL on a cloud warehouse: window functions, recursion, JSON.

Python for tooling, validators and tests.

Git, code review and CI as routine.

Reading an API’s documentation and probing the system before claiming anything about it.

Nice to have

HR or People Analytics data: headcount, turnover, compensation, recruiting.

BigQuery and Cube.dev in production.

Automate-dv or another Data Vault framework.

Row level security, policy tags, role based masking.

Claude Code or an equivalent agentic tool in your daily workflow.

Compensation

P75 of market Total Cash compensation (based on candidate seniority) + the opportunity to earn very significant equity in Zivo and become a partner over time.

Process

Hiring process

  1. 01

    Intro call

    30 minutes

  2. 02

    Deep-dive interviews

    2–3 sessions of 45 minutes

  3. 03

    Case Study

    Practical challenge

  4. 04

    Reference checks

    Conversations with former managers/peers

  5. 05

    Offer

    Final proposal

  6. 06

    Your future manager

    Leonardo Hahn

  7. 07

    Software Engineer | Founding Team

Know someone amazing?

Refer an A-Player and earn a R$10,000 bonus if they’re hired and stay for 3+ months.

Refer someone