# Jo Saillier, AI engineer

> Jo Saillier is an AI engineer who trains teams to use AI on real work, builds AI agents and apps for production, and tests them for security holes.

I'm Jo, an AI engineer who came in from the business side. After audit and banking consulting, I led the team behind a major Belgian bank's digital assistant and got hooked on what AI can do inside a company. Since 2024 I've built AI products, from voice agents and chat-based insights on large datasets to security research. Now I bring that experience to companies: I train your teams, build the agents they need and test them for security holes before they ship.

Languages: English, Dutch and French. Remote or on site.
Contact: https://joransaillier.com/#contact · LinkedIn: https://www.linkedin.com/in/joran-saillier/ · GitHub: https://github.com/jsr1234567

## Services

### Train: Get your team fluent in AI

In-person training shaped around how your team works, so people use AI on their real tasks the next morning.

Most teams don't lack AI tools. They lack someone who sits next to them and makes it work on their own files, their own clients, their own week. That is what this training does, in person, team by team.

**Track record**

- A large Southeast Asian bank: trained analysts to query their data in plain language through Looker, on a governed semantic model where every metric is defined once.
- In-person trainings where people with no coding background left with their own knowledge base, connected to their company's information.
- An AI lead-generation workflow that finds commercial real estate clients in China, for a corporate client.
- A major Belgian bank: grew its digital-assistant team from 4 to 14.

#### AI enablement

An adoption plan, in-person training per team and follow-through until it sticks.

Buying licences is the easy part. The hard part is the months after, when a few people try the tools and everyone else carries on as before. Enablement is that part.

**How it runs**

It starts by mapping where AI saves real hours in your teams and agreeing the tools and usage rules with IT and compliance. Then each team is trained in person, with follow-up visits until the new way of working holds.

**You leave with**

An adoption plan your managers can follow, teams that use AI on their own work, and clear rules on what goes into which tool.

For companies that know AI matters but don't know where to start, or that started and saw it fizzle out.

#### AI training for teams

Hands-on sessions over several weeks, built around each team's real work.

One session gives people ideas. Habits take repetition on real work, so this runs over several weeks.

**How it runs**

Short in-person sessions per team, on your own files, clients and processes. People practise between sessions and come back with real questions. No coding experience needed.

**You leave with**

People who use AI on their daily tasks, each with their own knowledge base connected to your company's information.

For teams where only one person uses AI today.

#### Workshops and talks

A hands-on workshop for a small group, or a practical talk on building with agents.

**Workshops**

Small groups, real work, no slide decks. People bring a task from their week and we build an AI workflow for it on the spot, so everyone leaves with something that works the next morning.

**Talks**

Practical talks on building with AI agents and using them safely at work.

**Languages**

English, Dutch or French.

For teams, company events and conferences.

### Build: Agents and apps that ship

From a working prototype to production: agents for your staff and your customers, model-agnostic by design.

In production, tools fail and inputs get messy. These agents are made for that: the model proposes, deterministic code checks and acts, and a person signs off on anything that matters. Evals, automated tests of the agent's work, run on every change, so imperfections are captured before your customers see them.

**Track record**

- Vox4AI: phone agents in four languages.
- A major Belgian bank: 80,000 conversations a month on its digital assistant.
- ĐiĐiRide: rider, driver, fleet and admin apps, in public beta on iOS and Android.
- Second brain: a knowledge base that AI agents maintain, in daily use.

#### Company agents, inside and out

Agents for your staff, and chat or phone agents for your customers, wired into your own tools.

**What**

Agents that do real work inside your company: assistants that handle internal requests for your staff, and chat or phone agents that answer your customers. They work in the tools you already run, and speak Dutch, French, English or German.

**How**

We start with one process that has a clear owner and a known cost, and automate it with AI agents in stages:

1. **Supervised:** a person validates every step of the workflow.
1. **Checked:** the workflow runs end to end, and a person verifies the result before it is accepted.
1. **Autonomous:** the workflow runs on its own, with continuous randomised test runs that a person reviews.

Evals score every change along the way, so imperfections are captured and fixed. Every release ships as code, with a rollback ready.

**Why**

When a tool fails, the agent says so instead of inventing success. Your systems stay the source of truth, not the model, so every outcome can be traced and explained.

For companies with work a machine should be doing, or customers waiting on a phone line.

#### Agent-ready knowledge base

One strong model working from your own knowledge base, instead of a stack of separate agents.

**What**

ICM, or Interpretable Context Methodology, organises your company knowledge so people and AI agents can both find their way in it. Instead of a separate agent per task, one strong model works from that knowledge base and follows the same structure your people do.

**How**

- **Structure:** folders and documents laid out the way your teams already work.
- **Routing notes:** short files that tell any agent where things live and what to read first.
- **Boundaries:** what agents may read, what they may change, and what never leaves.
- **Skills:** written procedures the agent follows for recurring work.

**Why**

Your knowledge stays in plain files you own, readable by people and easy to audit. You can switch model or provider without reconfiguring anything. And as models improve, one model on good context gets better, while a web of separate agents only gets harder to maintain.

For teams with scattered docs who want AI to work from what they know.

#### A personal agent, keys stay yours

A private agent for a leader or key person, on your own tools and channels.

**What you get**

An agent that works for one person. It handles mail, calendar and documents, runs scheduled tasks and reports back on the channel they already use. It runs on infrastructure you control and keeps its own knowledge base. It sees the names of your keys, never the values.

**How it works**

1. We list the recurring work worth handing off.
1. The agent is set up with its channels, memory, skills and schedules.
1. A week of real use, then a round of fixes and tuning.

For executives and key staff whose week is full of small, repeatable work.

#### From prototype to production

Web and mobile apps ready for real users, with a model only where it earns its place.

**What you get**

A working web or mobile app in your users' hands, not a demo. AI goes in only where it beats plain code, and the code, documentation and deploys are yours.

**How it works**

1. We agree the scope and what done means.
1. Short release cycles, tested early on real devices with real users.
1. Handover: code, documentation and one rule file that every coding agent on your team follows.

For teams that need it live, not just a demo.

### Secure: Checked before it ships

A second pair of eyes on your code and your running systems, with findings you can act on.

AI-written code runs. That doesn't make it safe: leaked keys, broken auth, an agent that does whatever a web page tells it to.

**Track record**

- Private disclosures in open source: 10 reporter credits across 6 projects, 276K GitHub stars combined.
- Named once fixed and public: Dokploy (Critical 9.9) and ToolJet (Medium 5.3).
- Live production assessments for private clients in Vietnam, through StateCtrl.

#### Software review (white box)

Your code and architecture checked from the inside, AI-built apps included, with a prioritised fix list.

**What**

A review of your source code and architecture from the inside. White box means the review sees what your developers see. Apps built with coding agents included.

**How**

- **Secrets:** keys and tokens in code, logs or the browser.
- **Access:** login, sessions and who can reach which data.
- **Agents:** prompt injection, and tools an agent should not have.
- **Architecture:** choices that will hurt at the next feature or the next audit.

Every finding is reproduced before it reaches you, ranked by severity, with the fix.

**Why**

Coding agents write code that runs, fast. They also skip the dull checks. A review before release costs far less than an incident after it.

For teams that built fast, often with coding agents.

#### Vulnerability assessment (black box)

Your live systems tested from the outside, the way an attacker would.

**What**

A test of your running systems from the outside. Black box means the test starts with what an attacker would have: no code and no inside access. Websites, APIs, apps and the servers behind them.

**How**

We agree the scope in writing first. Agents map your attack surface, everything reachable from the outside, and probe it for weak spots. Every finding is reproduced by hand, so nothing in the report is a guess. For each one you get the exact attack path, step by step, with a proof of concept that shows how the vulnerability is exploited, and a concrete suggestion for how to patch it. After your fix, a retest confirms the hole is closed.

**Why**

You see your systems the way an attacker does, before one does. Findings stay private, and nothing outside the agreed scope is touched.

For companies with systems already in production.

#### Extended security and compliance (with partners)

Your AI assistants checked for leaks and poisoning, and GRC and SOC compliance with partner firms.

**What**

Two things that go beyond a review. First, a check of your AI assistants, for your staff or for your customers: that they share only what they should (no other customers' data, internal documents, prompts or keys), and that they hold up against poisoning, where someone plants instructions or false information in what the assistant reads. Second, wider governance, risk and compliance (GRC) work and security operations (SOC) support.

**How**

Your assistants are tested together with a partner firm that specialises in chatbot poisoning, the way a curious user or an attacker would: crafted questions, instructions hidden in documents and web pages, and attempts to reach data outside a user's rights. Extended GRC and SOC compliance runs with StateCtrl's other partner firms, so one team covers the AI layer and the compliance around it.

**Why**

An assistant connected to company data can leak it in a single answer. Auditors and regulators now ask how your AI is controlled, under GDPR and the EU AI Act; this gives you the evidence to show them.

For companies running AI assistants on internal or customer data, or preparing for an audit.

## Selected work

- [Vox4AI](https://joransaillier.com/work/vox4ai) (Co-founder): Phone agents that answer calls in Dutch, French, English or German and book, move or cancel appointments.
- [StateCtrl](https://joransaillier.com/work/statectrl) (Founder): AI security company.
- [A second brain that AI agents maintain](https://joransaillier.com/work/second-brain): An Obsidian vault that AI agents file, link and search, from a Mac, a phone or a VPS.
- [ĐiĐiRide](https://joransaillier.com/work/didiride) (Team · Project owner · Project for Tasco JSC): Long-distance rides in Vietnam, with rider, driver, fleet and admin apps on iOS and Android.

## All projects

- [Vox4AI](https://joransaillier.com/work/vox4ai) (voice agents): Phone agents that answer calls in Dutch, French, English or German and book, move or cancel appointments.
- [StateCtrl](https://joransaillier.com/work/statectrl) (security): Finds the holes in AI agents and live systems, and shows you exactly how to close them.
- [A second brain that AI agents maintain](https://joransaillier.com/work/second-brain) (knowledge): An Obsidian vault that AI agents file, link and search, from a Mac, a phone or a VPS.
- [ĐiĐiRide](https://joransaillier.com/work/didiride) (mobility): Long-distance rides in Vietnam, with rider, driver, fleet and admin apps on iOS and Android.
- [Sesame](https://joransaillier.com/work/sesame) (security): Secures your keys and passwords on a hardware key, so a stolen laptop has nothing to give away.
- [Secrets Vault](https://joransaillier.com/work/secrets-vault) (security): Keeps API keys behind Touch ID, and gives each project only the keys you approve.
- [Agentic AI Build Week](https://joransaillier.com/work/aabw) (agents): Winners at Asia's largest agentic AI build week.
- [Cyber Watch](https://joransaillier.com/work/cyber-watch) (security): Weekly Telegram alerts on cyber incidents in your field, highlighting compromised codebases you use.
- [GreenOps](https://joransaillier.com/work/greenops) (optimisation): Carbon-aware scheduling for data warehouse jobs: waits out dirty hours, still meets deadlines.
- [Buinui](https://joransaillier.com/work/buinui) (marketplace): Adventure activities across Vietnam in one place, with an AI importer that drafts listings.
- [Daily AI Standup](https://joransaillier.com/work/standup) (showcase demo): A 9 am standup with 20 simulated agents, where a Chair agent picks what the manager must decide.

## Profile: From the business side

I started in audit, moved into banking consulting and ended up leading the team behind a bank's digital assistant. That taught me early that real users ask what you didn't plan for.

Since then, I've been building. If it doesn't work for the real user on the real device, it isn't done.

### Path

- 2026 to now: **Freelance**, and founder of StateCtrl · AI enablement, agents, security
- 2026: **Vox4AI** · co-founder
- 2025 to 2026: **Analytics engineer** · Joon Solutions, contract
- 2024 to 2025: **Vietnam** · business development, AI startup co-founder, data and AI engineering course
- 2022 to 2024: **Digital assistant lead** · major Belgian bank, via Avertim, Brussels
- 2020 to 2021: **Audit** · PwC Belgium
- 2015 to 2020: **MSc Business Administration** · KU Leuven

## Contact

- Contact section: https://joransaillier.com/#contact
- LinkedIn: https://www.linkedin.com/in/joran-saillier/
- GitHub: https://github.com/jsr1234567
