Twelve months ago, I left the safety of employment—not because I had a finished product, but because I had ideas I couldn’t pursue while working for someone else.
As someone who’s delivered dozens of platform implementations over the years, I’ve seen the same problems play out again and again—overly complex systems, clunky user experiences, and long, expensive rollout timelines.
Most AI tools today are no different. They're siloed, over-engineered, or limited to basic use cases. I didn’t want that. I wanted something simple, fast, and usable. Something that just works.
So I started building it myself.
When One Door Closed, AI Opened Another
Until a few years ago, I trained daily with an Ironman triathlon squad. I thrived on the routine, the discipline, the challenge.
Then health complications forced me to stop—suddenly and completely. No more training. No racing. No endorphin hit. It left a hole.
What do you do when the thing that fuels you disappears?
For me, the answer was simple: find a new mountain to climb. Something to master. Something to build.
The “Figure It Out” Mindset
Growing up, I watched my dad fix everything himself. Broken dishwasher? Fixed. Washing machine? Repaired. No callouts, no waiting—just figure it out.
That mindset stuck with me.
Whether it’s rebuilding a system architecture or debugging a chatbot, my instinct is the same: pull it apart, understand it, fix it, improve it.
So when I started thinking seriously about AI, I didn’t want to sit on the sidelines. I wanted to figure it out.
From Spreadsheets to AI Prototypes
It started with a personal pain point: tracking finances.
I’d built complex spreadsheets. Even prototyped a ServiceNow app (which I eventually lost in a PDI). But none of it worked the way I wanted.
So I asked: what if I used AI to automate this?
I built a simple flow to upload documents and extract data using LLMs. I embedded that into a vector store. I tested prompts like:
“Where did I spend the most money last month?”
“How much do I spend on bills?”
“What could I cut back on?”
In trying to solve my own problem, I accidentally discovered RAG—Retrieval-Augmented Generation. That lit a fire.
Then I Got Obsessed
The more I built, the more problems I uncovered. I didn’t want a toy chatbot—I wanted an agent that could:
- Understand context
- Use memory
- Trigger tools
- Give reliable answers
- And evolve over time
I built my own trace dashboard to see what was going on. I built a Tool Studio to define and test AI tools dynamically. I wired it all together with working chat and system integrations.
And I realised: I wasn’t just building a use case anymore.
I was building a platform.
Why I’m Building This Platform
Most of the AI tools I tried were either:
- Built for demos, not real work
- Locked to a single use case
- Full of hype, with no visibility into how they actually worked
They were siloed, rigid, and disconnected. So, I decided to build my own platform from the ground up.
A platform that:
- Onboards fast (10 minutes from invite to ServiceNow knowledge sync into vector store)
- Integrates easily with tools like ServiceNow, Azure, and your Teams environment
- Gives you visibility into what your agents are doing
- Scales from a single use case to full agent autonomy
- Doesn’t require a $100k implementation project to get started
No fluff. Just usable, adaptable AI—built on clean architecture, smart defaults, and real-world lessons.
What’s Coming First
The first release focuses on a single use case with big impact:
✅ AI Helpdesk Agent (inside Microsoft Teams)
✅ ServiceNow knowledge integration via RAG
✅ Optional Azure AD tools (create users, manage groups)
✅ Attachment reader that understands and summarises errors
✅ Tool Studio (internal use for now) to define AI tools with prompt + param logic
These features are already working and built on a secure, scalable architecture with audit logging, scoped access, and role-based memory boundaries.
Whilst most of these features are working, I’m taking my time refining access controls, role separation, and scoping. I’m not here to throw out a flaky MVP and call it done. I’m building for outcomes—measurable ones, like reducing ServiceNow backlog queues by up to 50%.
Behind the scenes, there’s also an additional capability quietly built for Service Owners. It’s not just about automating existing processes—it’s about reimagining them. This feature hints at a shift in how we structure service delivery, freeing teams from the rigid models of the past. I won’t reveal too much yet—but if you’ve ever thought, “there must be a better way,” you’ll want to keep an eye on what’s coming.
What’s Next
- I’m refining access controls and memory boundaries so agents can deliver value without admin access.
- A paid proof-of-concept will be available within the next 3 months.
- Full agent autonomy and multi-use-case support is targeted for end of year.
- And yes—Tool Studio will be opened up for customers once the platform has matured.
This is just the beginning.
If you’re serious about using AI to reduce queue time, automate intelligent actions, and integrate with platforms like ServiceNow without waiting 6 months to go live—follow along.
Next up, I’ll walk you through the platform, what’s working solid now, explain how AI can move from “cool demo” to “clear ROI.” and what’s coming in Phase 2. With screenshots.
No smoke and mirrors.
Because real AI? It’s not magic. It’s architecture, done right.
If you’re building something similar, or tired of the AI hype cycle—stay tuned. I’m not here to sell you a dream. I’m building the system I wish I had years ago.
Let’s figure it out.