Yaroslav Maxymovych

The AI Experts Catalog has been transformed into a 26-page bilingual resource, written to be cited by AI search engines and found by Google.

course.aiadvisoryboard.me/uk/promo↗

Outcome:26 pages in prod, fully static (instant opening); text and layout discrepancies are blocked by tests, not caught manually.

A person leaves a request on the website — and within a few minutes, they receive a call from an AI agent who conducts a conversation in Ukrainian with a live voice.

course.aiadvisoryboard.me/uk/promo/smartsim↗

Outcome:A live call by request was successfully completed (19 seconds of conversation). Along the way, two silent bugs were found, due to which some requests were lost without a trace: the dialer took the phone from the old client card instead of the number from the form, and the request card did not reach the manager at all if the name contained the symbol "&".

On all 500 blog posts, there was one button with an offer that suited a minority of readers; now each post leads to the one that is relevant to it.

aiadvisoryboard.me/blog↗

Outcome:148 articles now lead to the daily reporting tool, 352 to the corporate program; both paths have been tested on a live website.

A set of files where AI assistants leave their approved decisions, their mistakes, and who is currently working on what - expanded over 10 projects and made publicly available.

github.com/ymaxymovych/parallel-ai-dev↗

Outcome:Deployed on 10 projects. Filling is verified for honesty: all 215 references to real changes in the code exist, there are no fictional ones. The set is publicly available and can be used freely.

A curated selection of 14 long essays with full texts and dates - to prove to the editor in half a minute that the author writes regularly himself.

course.aiadvisoryboard.me/writing↗

Outcome:14 essays, about 93 thousand characters, live in prod with verified links to primary sources.

Instead of promises in the title, the visitor immediately receives a working tool: they enter the website address and the number of people - and see the structure of their company with hints on where AI will save time.

course.aiadvisoryboard.me/uk/ai↗

Outcome:The page is in production, works under advertising while preserving campaign tags; a short address leads to the desired language version, so advertising can be provided without long links.

Three ready-made procedures for an AI assistant: converting a dictated idea into a technical task, running it through three independent AI reviews, and checking whether what was ordered was actually done.

github.com/ymaxymovych/tz-skills↗

Outcome:The collection is used daily in our own development and has been publicly released as educational material.

The team's daily rhythm without meetings: each team member writes a brief report "yesterday / today / blockers", the system distributes tasks across the day's timeline and connects an AI advisor, and the manager sees the big picture across the entire team.

aiadvisoryboard.me/my-day↗

Outcome:The real client team uses it daily; the manager sees the team's status without morning calls.

Training teams to work with AI agents not "in theory", but under specific tasks of each participant: each participant takes away from the program the automation of their own workflow, not a lecture.

course.aiadvisoryboard.me/corporate↗

Outcome:On the 6th day, each participant had their first working automation; by the 7th day, almost half of the group (8 people) were demonstrating their own solutions. Each individual automation paid off the cost of all services. One participant did in two days what she had previously estimated would take three months. The best proof of value: one of the companies has already ordered training for a third group (they started with five people), and the classes consistently last 1.5–2.5 hours instead of the planned one hour – people don't want to leave.

The manager builds the structure of their company - departments and roles - in just a few minutes and sees in which roles AI will save the most time.

course.aiadvisoryboard.me/uk/orgchart↗

Outcome:The tool has been assembled and deployed (uploaded to the server); role edits and requests for in-depth analysis are collected in demand analytics - it is visible which specific tasks businesses want to automate. Public launch - pending the owner's decision.

A workflow where multiple AI assistants simultaneously perform different tasks in one product, without interfering with each other and delivering the results themselves.

github.com/ymaxymovych/parallel-ai-dev↗

Outcome:Dozens of changes per week reach the product without manual supervision of each step; there is an emergency switch that stops all automatic posting with one action.

For a manufacturing client, several AI models for recognizing text from photos of handwritten order forms were tested to select the most accurate one.

Outcome:Quantitative comparison of the accuracy of several models for a reasoned choice of solution.

Business call recordings are automatically converted into a structured knowledge base that can be searched by content, rather than scrolled through manually.

Outcome:830+ knowledge files processed and structured by AI instead of manual re-reading.

Call recordings with clients are automatically transcribed and analyzed by AI, which identifies the reasons for lost deals and provides hints to the manager.

Outcome:Automated evaluation and coaching on all recorded calls instead of selective manual listening.

Paid subscription for business clients: daily AI reports on the team, alerts about problematic patterns, and an interactive coaching chat.

Outcome:Paid feature with built-in cost control for each client.

The system itself collects news from the internet and Telegram, removes duplicates using AI, and publishes ready-made digests to the channel.

Outcome:Automatic daily publication of digests in a Telegram channel without manual work.

An automated conveyor (processing flow) that generates texts, images, videos, and voiceovers for social media on its own, with AI quality check before publication.

Outcome:Working conveyor at an early stage of launch, with built-in quality and cost control.

B2B lead search: the system finds potentially interested companies and itself writes personalized cold emails through AI, with protection against fabricated facts.

Outcome:Automated personalized letters instead of template-based mass mailings.

All meeting recordings are automatically transcribed, and any spoken phrase can be found by query - along with the moment it was said.

Outcome:First run on prod - 25 records out of 25 processed from disk, zero external quota spent.

Instead of manually checking 320 pages – three general rules and an automated test that doesn't let a broken page on a phone into the product.

Outcome:There were 15 issues - now there are 0. The Guardian paid off on the first day: it caught a cookie banner wider than the screen on every page of the site, which manual review did not see.

People who registered for webinars and exchanged promo codes had not been getting into the customer database for years - now they are getting in along with a draft of the first message.

Outcome:On live data: 22 client cards, 20 personal automation cards, 22 dossiers, 21 drafts, 0 automatically sent messages. Out of 19 drafts, 4 turned out to be unsuitable (non-existent link, kilobytes of garbage, hieroglyphics in the middle of Ukrainian text) — although the model's task directly prohibited this; now they are caught by mechanical verification after generation, rather than trusting the model.

Instead of manually compiling a commercial proposal, the system itself assembles a personalized page for a specific client: AI inserts their context, evidence from real cases, and tariffs. The manager only reviews it before sending.

Outcome:A personalized offer in minutes instead of hours of manual work; the system shows when the client opened the page.

The client is a logistics company specializing in international sea freight. In just one and a half hours, we designed and implemented the automation of the entire tender cycle with suppliers: collecting proposals, semi-automatic correspondence, conducting negotiations, and a mini-CRM system tailored to this process.

Outcome:Before: the manager was only able to negotiate with 10 out of 50 suppliers - she didn't have the patience for the rest. After: all 50 are now involved in the work. Approximately 40% of her work time has been freed up; the cost of the consultations paid for itself within the first three months, considering only the cost of her work time.

Landing page in three languages, Telegram bot and a simple request database for a small cleaning team that cleans short-term rental apartments.

Outcome:A working website with a domain, a bot, and an admin panel; the application automatically reaches the entire team, not just one person.

The page sells a sponsored spot in a blogger's giveaway: a five-step application, instant notification in Telegram, and a simple database of applications.

Outcome:Site in prod with automatic updates; requests are collected in one place instead of personal messages.