Alex Novak
Product lead on the Nordwind claims workbench and the Veyra support copilot, and the person who reads every enquiry that arrives here. Thirty-plus shipped products since 2016.
Eazetech was founded in 2016 as a frontend studio. Ten years later the same core team builds AI products, produces AI video ads and automates the operations underneath both.
Counts from our own project ledger, 2016 to 2026. Client outcome numbers on this page come from the client’s own systems and are documented in the case study each one links to. Clutch rating 4.9 from 23 reviews.
Eazetech started in 2016 with two people and one discipline: frontend. That was the work nobody else wanted to own properly — the interface, the states nobody had drawn, the performance problems that only show up on a real device. We took it because it was the part of the product a customer actually touches, and because being good at it was measurable in a way that broad claims are not.
The practice widened from there, and always for the same reason: a client asked for the next piece and we would rather build it than hand it to a stranger. Design came in-house in 2018 so the person drawing the screen was on the call when it broke. Backends followed, then mobile for Corva Bank in 2020, then a Web3 front end for Vaultic in 2021, then a security rebuild for MedArc in 2022 that taught us to treat any unverified output as a liability. None of it was a strategy on a slide. It was the pattern of saying yes to adjacent work and refusing to be bad at it.
In 2023 we put the first LLM features into products that were already live, and by 2024 the decision was explicit: AI-first, with automation as its own practice alongside it. What did not change is the standard the frontend years set. An AI feature that people route around has saved nobody anything, so the interface is part of the engineering, the output shows its sources, and the engagement is judged by what is running in production at the end of it. Ten years in, that is still the whole argument.
the company in six lines
One milestone a year, from the first client site to agents running in production. Where there is a case study, the number links to it.
Eazetech opens as a frontend studio. The first year is interfaces for other people's products: marketing sites, dashboards, and the unglamorous work of making slow pages fast.
A fintech client hands over an entire application interface rather than a page. The team reaches four, and the habit that still holds starts here: the person who designs the screen is on the call when it breaks.
We stop taking finished Figma files from agencies and start drawing the interface ourselves, because half of every estimate was going into translating someone else's mockup into something buildable.
An eight-week rebuild of a storefront checkout. The first engagement where the client measured the result in their own analytics and the number went on our own ledger.
Sixteen weeks, a full rebuild, and our first engagement at a scale where a bad release is a news story. Mobile becomes a permanent line rather than a favour.
Web3 work, taken on because the interface problems were interesting and the audit discipline was instructive. It taught us to treat an unverified output as a liability, which turned out to be the exact lesson AI needed two years later.
Twelve weeks rebuilding a healthcare platform's security posture. It set the practice we still run: least privilege, secrets out of code, and access that expires rather than lingers.
Not pilots. Existing clients get retrieval, extraction and drafting inside products that were already live, which is also where we learn how much of the work is evaluation rather than prompting.
The pivot is made explicit: AI products become the core, and workflow automation on n8n, Make, Zapier and custom pipelines becomes its own line with its own audit week.
Fourteen manual steps between an order and a dispatch become one trigger and four minutes. The same year, Quanta Realty goes from nine days of onboarding paperwork to one, and Arclight starts shipping 60 ad variants a month.
Nordwind decides claims 42% faster, Veyra closes 61% of tickets without a human, Bastion Health launches an AI care planner in eight weeks, and Helio Home passes 340 AI ads at 3.1x blended ROAS. Twenty-four AI products and agents are running in production.
Five people are the core team. Specialists join per engagement, but the names below are on every project, and the person who sells the work is the person who delivers it.
Product lead on the Nordwind claims workbench and the Veyra support copilot, and the person who reads every enquiry that arrives here. Thirty-plus shipped products since 2016.
Built the citation-first agent layer at Nordwind and the evaluation harness behind Veyra, where 61% of tickets now close without a human. Also wrote the matching logic in the Coastline run.
Designed the claims workbench that reached 95% adoption in its first week, and keeps the hook library that Helio Home's ad batches are scripted from.
Shipped 340 finished ads for Helio Home over twelve months on a weekly cadence, and 120 localised ad sets across six markets for Fairway Games.
Ran the MedArc rebuild that closed with zero critical findings, and sets up the client-account deployments and access expiry described on our security page.
We measure an engagement by what is running for real users at the end of it, not by what was demonstrated in the middle.
Bastion Health had an AI care planner live in 8 weeks, not a proof of concept in 8 weeks.
Any output a person has to trust carries its sources, and any number we publish names the system it came from.
A model that people route around has not saved anyone anything, so the interface is treated as part of the engineering, not a layer applied afterwards.
The Nordwind workbench hit 95% adoption in week one; the budgeted adoption fight never happened.
Everything runs in your accounts, under your ownership, with the documentation needed to change it without us.
The plain version, including the badges we do not have.
How client code, data and access are actually handled is written out in full on the security page — where things run, what the model endpoints retain, and when our access is removed.
Ten years, 120+ projects, five people who have shipped together. A thirty-minute scoping call answers fit and gives you a rough estimate, with a reply within one business day from Alex Novak.