# Coaching expertise, built into an AI coach

> An AI coach in text and voice, in several languages, built for TNM Coaching over about 18 months and two major versions. The hard part was not the model. It was turning 25 years of TNM's coaching method into behaviour people trust.

- Client: TNM Coaching
- Type: AI coach in text and voice, in several languages
- Category: AI product
- Duration: About 18 months, two major versions
- Status: Second major version
- Stack: Text and voice · Several languages · Memory across sessions · SMS and email accountability · Privacy-aware HR reporting
- Role: End to end, first call to launch and iteration
- Result: 95% of onboarded regular users completed a Leelou session (TNM's system-wide analytics snapshot, 27 August 2026. Not lifetime reach: inactive users may be deleted under TNM's GDPR retention.)
- Result: 45% package completion (TNM's system-wide analytics snapshot, 27 August 2026. Not lifetime reach: inactive users may be deleted under TNM's GDPR retention.)
- Result: 13 min average session (TNM's system-wide analytics snapshot, 27 August 2026. Not lifetime reach: inactive users may be deleted under TNM's GDPR retention.)
- See Leelou: https://tnmcoaching.com/business/leelou-ai-coach/
- Built by: Luka Abramović (https://lukaabramovic.com/)
- Canonical: https://lukaabramovic.com/work/leelou-ai-coach/

---

## The starting point

TNM Coaching has 25 years of coaching method behind it. Before Leelou, it had no AI product at all. TNM wanted one that could carry its method into a new format. Off-the-shelf chatbots could produce answers, but they couldn't coach the way TNM's coaches do.

## Coaching people trust, not a chatbot.

The hard part was turning 25 years of coaching method into behaviour people trust, in text and voice, in several languages, and across sessions rather than one conversation at a time.

### Method

Leelou had to coach the way TNM's coaches do, built around their method.

### Continuity

What happens in one session has to carry into the next, with accountability in between.

### Privacy

HR gets reporting, and the people being coached keep their privacy.

## My role

I led Leelou through two major versions over about 18 months. I learned the method from TNM's coaches, designed the conversation, memory and reporting, and tested with the coaches until Leelou met their standard. I ran it end to end, from the first call through discovery, scope, solution design and leading the build with the Adamant Code team, to launch and iteration.

1. First call
2. Discovery
3. Scope & proposal
4. Solution design
5. Build lead
6. Launch
7. Iteration

<VideoFigure src="/media/leelou-voice-demo.mp4" poster={leelouPoster} label="Leelou voice demo, 1 minute 37 seconds" tag="▶ Leelou voice demo · 1:37" />

## More than a chat window.

### Text and voice

People can coach with Leelou by typing or by talking.

### Several languages

Leelou coaches in more than one language.

### Memory across sessions

Each session builds on what came before, so people don't start over every time.

### SMS and email accountability

Follow-ups between sessions by SMS and email keep people on track.

### Privacy-aware HR reporting

Reporting for the organisation that respects the privacy of the person being coached.

## Coaching first, then voice and reporting.

Leelou shipped in two major versions over about 18 months. The first was text only. The second added voice, admin tools and HR analytics.

### Version 1

Text only, and focused on getting the coaching right. That meant Leelou nudging people toward their own thinking instead of handing them answers, and keeping coaching separate from direct advice. TNM's coaches kept testing it, rather than treating the first working conversation as finished.

### Version 2

Added voice, admin tools and HR analytics. People can move between text and voice while Leelou keeps the session's context and its coaching posture. HR analytics show aggregate engagement signals without exposing private coaching conversations.

## What TNM's analytics show.

In TNM's system-wide analytics snapshot of 27 August 2026, 95% of onboarded regular users had completed a Leelou session, package completion stood at 45%, and the average session lasted 13 minutes. Inactive users may be deleted under TNM's GDPR retention, so these figures describe that snapshot, not lifetime reach.

<MetricTiles
  size="m"
  items={[
    { value: '95%', note: 'Of onboarded regular users completed a session. TNM analytics snapshot, 27 August 2026; not lifetime reach.' },
    { value: '13 min', note: 'Average session. TNM analytics snapshot, 27 August 2026.' },
  ]}
/>

> You didn't just build features. You listened. The challenge was not only developing technology, but translating the human complexity of coaching into something usable.
>
> – Zoran Todorovic, Senior Partner, TNM Coaching

## Building believable AI coaches.

In January 2026 I joined TNM Unplugged, TNM Coaching's podcast, to talk about building Leelou: "Building Believable AI Coaches with Luka Abramovic".

<YouTubeFacade id="Foa_DwfMEXY" title="Building Believable AI Coaches with Luka Abramovic" poster={tnmThumbnail} />

## What this proves

An AI coach earns trust when it carries the method of the people who built the practice, and when those coaches test it until it meets their standard.