I love findingproblems& building thesolutions

I’ve built two startups, delivered projects for clients, and set up AI systems that do real work every day. Here’s what I made, how I did it, and how I can help.

Ali Shaverdi holding his grey and white cat, Yuki

I'm one of those people who wakes up at 4:30 am to work out, and I've done it for 16+ years. I can't count how many of those mornings I thought about how easy it would be to keep my eyes closed and go back to sleep. Every time, I decide it's not who I want to be. I want to be someone who keeps trying to achieve their goals and puts in the work to get there.

It taught me grit. No matter how many mistakes I make or how many times I fail, I keep trying. I won't say that if you keep trying, you'll eventually succeed. I always have two options: keep trying, or don't.

If we keep trying, we may or may not succeed. But if we don't, we will never succeed.

Say your mom's birthday is coming up and she lives in another country. You can ship the gift yourself, but then you're paying for shipping and waiting on delivery and customs. Or you can try to find a shop there yourself, site by site, in a language you might not read, and hope the payment goes through. And I kept thinking, okay, the right shop is already in her country, so why does the gift have to travel at all? So we built Flomaru. You pick the country, you browse its local shops in your language, you pay in your currency, and the shop delivers it there.

The whole Flomaru home page, top to bottom: the hero with a city picker, three reasons to trust it, local products, how it works, reviews, local shops, common questions and the footer

The Flomaru website: pick the city, and the local shops there show up. Scroll inside the picture to see the whole page.

My AI work was spread across notes, scheduled jobs and automations that never talked to each other. So I thought: what if I can visually see the exact step-by-step process of all the different things that the AI is doing? Not only that, what if I can, in real time, see exactly what it's doing, when it's doing it, and be able to just visually keep track? Maybe that is the way that we should be managing our AI work.

Portfolio · BoardsHow Vizi works
Saved
Make plan

How Vizi works. The steps in order:

  1. Start: Describe the work (Owner). The owner describes the work they want done. Everything starts with a person, so the plan reflects what they actually need.
  2. Draw plan on board. The plan is drawn on a visual board with typed shapes. One item can show up in many views without copies, so everyone's looking at the same thing.
  3. Agent proposes changes (Claude). An agent proposes changes to the plan. They stay separate from the board until a person accepts them, so nothing changes behind anyone's back.
  4. Decision: Person approves? (Owner). Only a person approves. Agents ask for every permission up front so they don't stop halfway, and anything turned down goes back to the agent to rework.
  5. Agents carry out plan. The approved plan becomes work the agents carry out. People connect their own AI from outside, so they use the agents they already have.
  6. End: Work visible to all. The work, and who's doing what, stays visible to people and agents alike. I built it because I couldn't see who was doing what.

How Vizi works, drawn on a Vizi board. The agent proposes, a person approves.

Maki

A lot of our sellers on Flomaru make really good products, but the only pictures they have are the ones they took on their phone. So I thought, okay, what if they just send us that one picture, and we give them back everything else? The studio shot, the lifestyle shot, the one for a birthday. The only rule: it still has to be their actual product.

The Maki app home screen: the seller's own phone photo of a flower and cake gift box beside four new product images made from it, waiting for review

Maki: the seller's own phone photo, and the images made from it.

Maru

Every new AI chat shouldn't need a fresh explanation of my company, my projects and what it's allowed to do. And I was explaining it every single day. So the idea was simple: the files are the brain, and any AI is just a worker. A new chat reads the map of the work first, then gets on with the job.

Maru as an always-on assistant: a dial labelled Maru holding context, permissions, decisions and unfinished work, directing Claude and Codex, with a live feed for Ali: morning brief ready, 2 decisions need you, handover sent to Codex.

Loci

A sentence can be translated correctly and still sound wrong for the person reading it. Native speakers kept telling me a correct word-for-word marketing line still read like a translation. So Loci captures what the sentence is supposed to say first, then writes it for that market, and a person approves the exact version.

HelloHolaمرحبا你好नमस्ते

Coni

A newer B2B company can have something useful and still struggle to reach the person who can move the conversation forward. Some startups have a product that's close to a guaranteed close and still spend most of their time cold calling. So Coni skips all that: one useful call with the right person, not a long list of attempts.

Your offer Champion Decision maker

AI systems

Skills, workflows and automations I built so AI does real work for me every day. Here are four of them, and the board I drew for each one.

Two AIs, one checks the other

An AI that builds something will often tell you it's done, and I kept seeing “done” mean “I changed something”, not “it works”. So I had a different AI review it, against checks written for that change, and if it fails it goes back to the builder. A person still signs off.

Portfolio · BoardsTwo AIs, one checks the other
Saved
Make plan

Two AIs, one checks the other. The steps in order:

  1. Start: Task assigned. Every task starts with what done should look like, written down. That's what the review checks against later.
  2. One AI builds (Claude). One AI does the work and writes down what it changed. It doesn’t get to just say it’s done.
  3. A different AI reviews (Codex). A different AI checks the work on its own, against the checks written for this change. It’s not the builder grading itself, so it doesn’t share the same blind spots.
  4. Decision: Passes review?. If the reviewer finds a problem, the work goes back to the builder to fix. Nothing moves on a promise.
  5. Person approves (Owner). A person looks at it and signs off. The AIs can check each other, but they don’t get the final say.
  6. End: Done and recorded. It’s marked done, with a record of what changed and who approved it. A daily brief tells my co-founder what’s live and what isn’t.
  7. Note: Nothing is signed off by the AIs alone.. This is the one rule the whole thing hangs on. Two AIs catch more than one, but the call stays with a person.

Testing a way of working before trusting it

A month of design work with AI kept failing in the same ways, and I was guessing at fixes, changing three things at once. Then I wrote one rule down: find the causes from evidence. The skill runs that rule the same way every time. I planted defects on purpose to test it, and a checker with no AI in it found 93% of them in 13 seconds, with no false alarms.

Skills · PortfolioTesting a way of working
Saved
Use it
The picture of the skill’s instructions. Edit either one; the other updates, and a person reviews the change.Instructions

Testing a way of working. The steps in order:

  1. Start: Repeated AI failure. It starts when AI keeps failing the same way. Instead of guessing at fixes, I go looking for the cause.
  2. Diagnose from chat records. I go back through the chat records and diagnose in rounds. The fix has to come from evidence, not a hunch.
  3. Write hypotheses. Each possible fix becomes a hypothesis, with its pass line written down before any test runs. That way I can't move the goalposts after I see the result.
  4. Test with blind reviewers (Claude). Fresh chats do the work and blind reviewers score it against answer keys they can't see. Anything a plain program can check gets moved out of the AI step.
  5. Decision: Every hypothesis judged?. I keep going until every hypothesis has a verdict, and "not reproduced" counts as one. Those stay in the log, because a method that only reports wins isn't worth much.
  6. End: Method adopted or dropped. The fix is adopted or dropped based on what the tests showed. The methods that make it now sit underneath my other design and build processes.
  7. Change one thing. If a hypothesis is still open, I change one thing and test again. Changing three things at once means never knowing which one helped.

Build to Match the Design

Wiring a design to real data is scary, and not because of the look. It's losing the look while the data goes in, and calling stuff “done” when it isn't. The approved design comes in unchanged, one worker takes one item, and a blind auditor checks it, round after round. On Vizi that was 547 screens, and all 547 matched the board.

Plans · PortfolioBuilding to match a design
Saved
Make plan

Building to match a design. The steps in order:

  1. Bring in and measure. The approved design comes in unchanged and I measure each part of it. From here on the design files are frozen, so nobody edits the look just to make a check pass.
  2. Build shared data first. The shared data and save pieces get built before any single item, with one owner per file. That way the workers building items don't trip over each other.
  3. Build an item (Claude). One worker takes one item and wires in real data and actions, with scripted user journeys running first. It never touches the frozen design, and design questions come to me in a batch.
  4. Build an item (Claude). Items get built side by side, one worker each. When an audit sends an item back, this is where it gets fixed, under the same rules.
  5. Audit each item (Claude). A separate auditor checks each item without seeing how it was built. It's not the builder marking its own work, so it catches what the builder missed.
  6. Decision: Zero issues?. If the audit finds anything, the item goes back to be fixed and audited again. The loop only ends when the checks find no issues.
  7. End: Item accepted. The item is accepted, with its screens matching the approved design. Then it's ready to hand over to my technical co-founder.

The Live Data Copy

My dashboards disagreed with reality. Web analytics said zero purchase events, and the payment processor only sees payments that go through, so it misses cancellations and abandoned checkouts. The fix was a live copy of the production database, and that's now the source of truth for orders. It also showed me three traps, like revenue added across currencies as if it was all one.

Map · PortfolioOne live copy of the data
Saved
Make plan

One live copy of the data. The steps in order:

  1. End: Production database. This is where real orders live, including cancellations, orders paid another way and abandoned checkouts. Web analytics and the payment side only ever see part of that.
  2. Continuous copy. A copy of the production database sits in a warehouse and changes as the live one changes. It's always on, so every report reads the same up-to-date records.
  3. One definition of an order. An order is defined once, in one place, and every report uses that definition. Orders, revenue and attribution all get counted the same way.
  4. Dashboards. Every dashboard takes orders, revenue and attribution from the copy. That includes the marketing and sales dashboard I share with my co-founder and the monthly investor report.
  5. Other analytics tool. Web analytics only covers the part of the funnel it can actually see. Orders and revenue come from the copy, not from here.
  6. Note: If tools disagree, the copy wins.. When a tool's numbers don't match the copy, the copy wins and the gap has to be explained. That's how it caught revenue being added across currencies and test traffic inflating checkouts.

There are more.

Every one has its own page: the problem I was fixing, how it works and what it costs to run.

Projects

Here's what I built for a client, what I did on contract, and the client and research projects.

Someone needed a team to build a marketplace. Nobody had asked me.

Harvest Network marketplace

I built this for a client. Someone was looking for a team to build a marketplace, so I reached out and won them as a client, and they paid. I scoped it first so we both knew what was in and what was out. Then I put a team together, ran the build and delivered the whole thing.

Team project. I led it.

Rogers wanted to figure out what else they could do, outside of connectivity.

Rogers innovation consulting

This was contract work. Rogers brought me in to figure out what else they could do outside of connectivity. I can't say what the problem was or what came out of it, because that's Rogers' to share. The work went the way most consulting does: research, then ideas, then proof, then a deck. I wrote the hypotheses, designed the interview guide and the survey, ran the interviews, and built the deck.

Innovation consultant

I was the person Canada Post talked to.

Canada Post consulting project

Canada Post was our client, and our team worked with them directly. We agreed the scope, did the research and handed over a report they could use. I was the client liaison and the interviewer, and I ran the sessions with them. I can't say what the project was about or what we found, because that's Canada Post's to tell.

Team project. I led the client sessions.

There are more to checkout

Five ways I can help.

Make the work easier to run.

I map the work, separate what needs judgment from what can be repeated, then build the tools around it.

Work through a product problem.

I get a usable version in front of people early, then turn the decisions into a prototype or a clear build handoff.

Find out what the business actually needs.

Sometimes the answer is a build and sometimes it isn't. I research the problem before I recommend the next step.

Connect the product to its customers and partners.

I work through how people find the offer, understand it and get started.

Teach your team to build with AI.

I show your team how I plan the work, hand it to AI and check the result, so they can run it themselves.

How I work

If I do it once, I shouldn't do it twice.

That's the rule I work by. When I finish a piece of work I break it into its smallest steps and experiment with automating them, one at a time.

With AI, there's no such thing as a good result. There are consistent results.

A system that's brilliant once and wrong twice is worthless. So nothing I build gets trusted on day one.

I plan by day. The systems ship by night.

My work is half directing AI and half doing the things AI can't. I break the work into pieces, file them where my agents pick them up, and by morning it's done and reported back.

More about me

Want to tell me what challenges you're facing?