Two People Ops tools I built, both live and clickable

Five years leading People Operations at Kialo Inc., reporting to the CEO. Since leaving I completed AIHR’s AI for HR Specialist certificate and built two tools. Both are working builds, not mockups: the links below open the real thing.

contact.tobiaswalter@gmail.com · LinkedIn

The track record behind them

Up to 30 people across ten countries, for five years.

• Redesigned onboarding, roughly 25% faster ramp-up. Built around new hire feedback. The same problem the onboarding agent now automates.
• Traced a turnover spike to its root cause. Looked for what the people leaving had in common, found a management-level cause, raised it with leadership, and turnover came down.
• Negotiated 30% off our EOR contract. Deel, for the company’s first US employment contract.
• Turned a CEO’s security concern into working policy. A signed policy, a tracking system, and a Slack check-off. Getting compliance was the easy part; building the understanding behind it is what made it self-sustaining within months.
• Ran the entire employee lifecycle: hiring, onboarding, payroll, performance, employee relations and offboarding.
• Neither tool below is production-ready. Each says plainly what is built and what is not.

Running the same checklist by hand, person after person

Onboarding by hand is inconsistent by nature, hard to scale and time intensive.

• There is a second problem underneath it. Nobody wants to look unprepared in front of people they just met, so some stay quiet and guess. The manager finds out when something goes wrong.
• So the target isn’t only speed. An agent removes the social cost of asking the question that feels too small to bring to a person.
• And I wanted real understanding, not box-checking. A checklist marked "done" doesn’t mean the material landed, so every handbook ends in a quiz.

How it feels is one half. How it sounds is the other.

Words build culture. Agree the few that stand for yours, and the agent repeats them at every step rather than stating them once and never again.

The quizzes test values, not just process. TaskTerminator, the fictional company this demo runs on, has five:

New hires move forward when they’re ready, not on a fixed schedule

• Nothing unlocks on a date. A phase opens only when the one before it is finished. Nobody is pushed forward or held back.
• Five people involved: the new hire talks to Margaux; HR/Ops schedules and takes escalations; mentor and supervisor handle the human conversations; a peer group covers the cohort call.

I wrote the four handbooks and 30 quiz questions behind Phase 1, each tagged to its source section so a wrong answer points back to the right paragraph.

Four things the agent deliberately will not do

How it was built: the flow is hand-built in Botpress Studio, no code. Claude Code drafted the methodology, structured content, and debugged live. Knowing which half to hand to AI was most of the job.

A quiz answer going wrong, and being caught

• Runs in a real Slack workspace and as a standalone web chat. Same flow, both channels.
• The tone is calibrated to the moment, not to the company. Reassurance is right for someone nervous in week one. It is not a statement of company values, and it shouldn’t be: ownership comes later, once the ground is solid.
• Answer one wrong on purpose. It is the quickest way to watch Margaux catch it and point you back to the right section.

What it is, and what it is not yet

• Every new hire hears the same company. Mission and values are taught and tested, not assumed, and the wording doesn’t drift with whoever is running onboarding that week.
• Designed to extend to offboarding on the same checklist-automation logic, though that is not yet built.

The same first two weeks, two ways

A wizard that builds the sequence itself

• Whoever owns onboarding uploads the handbooks and sets the learning volume: a quick orientation, or a thorough grounding.
• The tool splits the docs into chunks and drafts a quiz for each, on the same 100%-pass rule already built and tested in Phase 1.
• Every generated question needs a human review before it goes live: approve, edit, or discard. Publishing stays blocked until that’s done.
• The sequencing would be custom-coded, since no chatbot platform builds a flow for you. The wizard feeds approved content into a live sequence through the platform’s API.
• Deliberately content-only. It doesn’t guess at org structure or mentors. The relational side stays a human decision.

Putting together a dashboard is easy. Making it meaningful isn’t.

One reader, 90 seconds, no analyst in between. Every decision below follows from that.

• Who it’s for: the founder or CEO of a ~100–250 person company, reading it directly rather than via an analyst.
• Why that size: below ~50 a founder knows everyone personally; past ~100 that stops scaling and aggregate signals start paying off.
• The bar: three questions in 90 seconds. Am I losing people I can’t afford to lose, and where. Is the team more or less productive per dollar. What changed since I last looked.
• An honest limit: my own CEO’s real questions often weren’t the kind a dashboard answers well.

One page, governed by three rules

• One metric, one home. Nothing appears twice. Retention rate was cut as redundant: it was literally 100 minus turnover.
• Exception-based. Healthy numbers stay quiet, problems get loud. Sick leave renders as a single quiet line and only enters the hero row when it actually spikes.
• Answer one of the three questions, or don’t earn the space. Four hero cards carry them, color-banded so the row reads by color before a single number is.

How it was built: I had Claude critique each version as a busy, skeptical CEO, then rebuilt around the question that survived: does the reader understand every number in one line, without digging?

The hero row, doing its job

• Illustrative data, not a live company. A ~140-person SaaS business with regretted attrition concentrated in Customer Success, rising over the last four months.
• Every warning state is showing on load. That’s deliberate: you can watch the page get loud about a real problem instead of taking my word for it.

What changes for the CEO

• A meeting only when something fires. The page carries the routine reporting and shows what has gone red at a glance, so a conversation is reserved for that.
• Team-level problems surface. A healthy-looking company average turns out to be one department with a worsening trend.
• The page states what it can’t tell you. Four limits sit on the dashboard itself, not in a footnote nobody opens, including turnover measured against today’s headcount and an attrition cost that is only an estimate.

Every metric is attached to a decision

A metric with no decision behind it is a vanity KPI. Three from the live build:

The same logic drove the exclusions: promotions and internal mobility were cut entirely, because a monthly count at this scale is a handful of events with no threshold and no action attached.

What building these actually demonstrated