Illustration: Maxim smiles and hugs two characters. One has the Claude logo as a head, the other has the n8n logo.

AI Solution Engineer

I am Maxim Vicneanschi.

I help companies grow with artificial intelligence. I automate processes and build AI agents, so that employees can focus on their main work.

Experience in numbers

I own each automation from the first talk with the business to release and support.

Cases

25 AI systems I built at work, and one app of my own. Each card shows the result and the process: where AI works, where rules work, and where a person decides.

Finance · AI automation agency

AI driver payroll

Weekly driver payroll for a US logistics company, calculated by AI agents.

  • Team: 16 → 2
  • 99.7% accuracy on 5,000+ orders in a 4-month test
  • Run time 45 → 16 s

How it works

  1. StartTrips and orders come from the order database.
  2. AIAI agents read and sort each trip.
  3. RulesThe system checks driver and broker data and calculates pay.
  4. PersonA supervisor gets each item that does not match.
  5. ResultDriver statements go out.

4 days → 22 min

weekly payroll

Before
16 accountants spent 4 days every week matching trip data and statements by hand for 100 drivers.
My part
I did it alone: process mapping, data flow, agent prompts, exception handling, testing and handover.
Tools
n8n, LLM agents, order database

Sales · B2B marketing services company

AI call coaching

Each customer call gets a score and coaching notes for the rep.

  • 420+ calls graded

How it works

  1. StartThe call recording and its transcript arrive after the call.
  2. AIAI finds the call type and scores the call against the team's own rubric. Each score quotes the call.
  3. AIAI lists the missed opportunities and writes one clear tip.
  4. ResultThe rep gets a scorecard. The team lead gets a team report.

60 → ~8 min

to give feedback on a call

Before
Team leads could review only a few calls. One review took about an hour, so most reps got feedback late or never.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, Claude, Gemini, Supabase
Size
12 workflows, 186 nodes

Operations · AI automation agency

AI support agent

An AI agent answers an e-commerce brand's customers in minutes.

  • 70% resolved at once
  • 1,200+ users a day
  • 15,000+ notifications a month

How it works

  1. StartA customer asks a question.
  2. AIAI finds the answer in the order data and replies.
  3. PersonThe support team gets each question that AI cannot close.
  4. RulesNotifications reach customers before they ask.

24 h → <5 min

response time

Before
Customers waited up to 24 hours. The team answered the same questions every day.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, LLM API, order system, messaging

Marketing · B2B marketing services company

1-to-1 landing page generator

A personalised landing page for each target account in an ABM campaign: visuals, logo, video, insights and a call to action.

  • ~50 pages a day
  • 5 sub-workflows run in parallel
  • Logo cache, model fallback and strict output checks

How it works

  1. StartThe campaign platform sends a webhook with the target company's domain and the source PDF report.
  2. RulesThe system checks the input and pulls the company's branding.
  3. AIVisuals: Gemini reads the PDF cover, recomposes it, builds report covers and a background, and picks an accent colour that passes WCAG contrast.
  4. AILogo: the system checks its cache, then logo APIs and image search. Gemini compares the candidates, picks the best one and fixes it when needed.
  5. AIVideo: the system finds the company's YouTube channel. AI picks the best video and checks that it fits a business page.
  6. AICopy: Gemini writes the insights and the CTA banner. A second model takes over when the first one fails, and code rejects copy that breaks the length rules.
  7. RulesCode merges the five results into one page config and sends it back to the platform.
  8. PersonA marketer reviews the page and publishes it.

3 more steps in Details.

30 → 5 min

to produce a page

Before
Account-based campaigns need one page per account. One page by hand took about 30 minutes.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, Gemini (text, image and vision), OpenRouter (fallback model), PDF.co, Supabase, Brandfetch, Logo.dev, SerpAPI, Tavily, YouTube Data API, Google Drive
Size
6 workflows, 157 nodes

Finance · AI automation agency

AI invoice processing

AI reads each invoice, matches it to the purchase order and posts it. Nobody types.

  • ~95% need no human input
  • Team: 4 → 1

How it works

  1. StartAn invoice arrives as a PDF or a scan.
  2. AIAI reads the invoice data.
  3. RulesThe system matches the data to the purchase order, contract and payment terms.
  4. PersonA finance officer gets each exception with an explanation.
  5. ResultThe system posts the invoice for payment.

5-10 days → <1 day

to process an invoice

Before
A team of 4 keyed in invoices and checked each one against its purchase order.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, OCR, LLM extraction, accounting system

Marketing · B2B marketing services company

AI-search visibility audit

Shows how AI assistants describe a brand next to its competitors.

  • Report by email

How it works

  1. StartA requester enters the client domain and competitors in a form.
  2. AIAI reads the client's website and builds a brand profile.
  3. RulesThe same buyer questions go to 4 AI assistants.
  4. AIAI classifies the answers, reads the cited sources and writes a gap analysis.
  5. ResultA report arrives by email.

4 AI assistants

asked the same buyer questions

Before
Nobody knew what AI assistants tell buyers about the brand and its competitors.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, LLM APIs, web scraping
Size
5 workflows, 142 nodes

Marketing · B2B marketing services company

ABM report check

Checks an ABM report against its source files before it goes to the client.

  • Runs before each delivery

How it works

  1. StartAn analyst uploads the report and its source files.
  2. RulesThe system reads all files at the same time.
  3. RulesRules check each case in the report against the source data.
  4. ResultA verdict for each case and a list of accounts to fix.

~11 s

to check a report

Before
Analysts compared the report with the source files by hand.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, JavaScript
Size
1 workflow, 24 nodes

Sales · B2B marketing services company

Client dossier

One page for each client, found by email. It holds every call, message and other contact, with every offer the client got, liked or declined.

  • Fills itself after each call, email or message
  • Feeds the follow-up drafts
  • Stops repeated offers

How it works

  1. StartA call, an email or any other contact with a client takes place.
  2. AIAI extracts what we offered, what interested the client and what they declined.
  3. RulesThe system finds the client's page by email and adds the new facts to it.
  4. ResultThe team sees the whole client history in one place.
  5. RulesBefore a follow-up draft, the system checks the dossier, so we never offer the same thing twice or bring back an offer the client declined.

Every contact

with a client in one profile

Before
The client history was spread across call notes, email and chats. A rep could offer something the client had already declined.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, LLM APIs, Supabase
Size
8 workflows, 102 nodes

Sales · B2B marketing services company

Follow-up email drafts

After each call the rep gets a ready draft of the follow-up email.

  • Uses the call analysis and the client dossier
  • Rep rates each draft

How it works

  1. StartThe call analysis from call coaching is ready.
  2. RulesThe system pulls the client dossier: every past offer, what the client liked and what they declined.
  3. AIAI drafts the email from the facts of the call. It marks missing facts as placeholders.
  4. RulesThe system checks the draft before the rep sees it. A draft never repeats an offer the client already declined.
  5. PersonThe rep edits and sends the email. The system never writes to the client.

45-60 min → 25-40 s

to draft a follow-up email

Before
Reps wrote each follow-up email by hand. One email took 45 to 60 minutes.
My part
I built the whole system alone. I turned the CS lead's objections into the rules that check every draft.
Tools
n8n, Claude, Gemini, Supabase, email
Size
2 workflows, 46 nodes

Sales · B2B marketing services company

Meeting-prep briefings

Reps get a short briefing on the client before each meeting.

  • Briefing on demand

How it works

  1. StartEvery night the system reads tomorrow's meetings from rep calendars.
  2. AIAI agents research each company and each attendee.
  3. AIAI turns the research into a one-page briefing.
  4. ResultThe briefing arrives by email before the meeting. A rep can also ask for one on demand.

~50 meetings

prepared every night

Before
Reps researched each company and attendee by hand. With a full calendar they often skipped it.
My part
I built the whole system alone and improved it from rep interviews. Reps rate the briefings 7.5/10.
Tools
n8n, LLM research agents, calendar and email APIs
Size
14 workflows, 243 nodes

Sales · B2B marketing services company

Outreach engine

A personal message sequence for each contact in a target account.

  • Lead prioritisation: 8 workflows → 1

How it works

  1. StartSales submits a list of prospects.
  2. AIAI builds a profile of each company and each person.
  3. AIAI writes 6-7 LinkedIn messages and emails with send dates.
  4. PersonEach day the messages due today become drafts. The rep sends them.

6-7 messages

written for each contact

Before
Personal outreach meant manual work for each account. Lead prioritisation was spread across 8 workflows.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, LLM APIs, CRM data
Size
6 workflows, 137 nodes

Sales · B2B marketing services company

Competitor intelligence

Shows sales which prospects already engage with competitors.

  • Request through a web form

How it works

  1. StartSales uploads competitor reps' LinkedIn profiles and a date range in a form.
  2. RulesThe pipeline collects their reactions and comments.
  3. AIAI checks each engaged person against the Ideal Customer Profile (ICP) and scores the person Hot, Warm or Low.
  4. ResultA spreadsheet report arrives by email.

~4 min → 50-70 s

per run

Before
Nobody could track by hand who engages with competitors on LinkedIn.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, Apify, LLM matching
Size
11 workflows, 150 nodes

Sales · B2B marketing services company

Contact list check

Checks a contact list against live LinkedIn data.

  • Status written back to the sheet

How it works

  1. StartA contact list arrives in Google Sheets.
  2. RulesThe pipeline collects each person's current LinkedIn profile.
  3. RulesMatching rules check that the person still works at the company.
  4. ResultA status for each contact goes back into the sheet.

Every contact

rechecked against live LinkedIn data

Before
People checked each contact on LinkedIn by hand.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
n8n, Google Sheets
Size
2 workflows, 45 nodes

Knowledge · B2B marketing services company

AI knowledge bases

A manager asks a question and gets an answer with sources.

  • 8 departments, one base each

How it works

  1. StartJira tickets and Confluence pages become files under version control.
  2. AIA Claude Code agent searches the base.
  3. ResultThe manager gets an answer with dates and links to the sources.
  4. RulesA daily check flags new material to add.

<2 min

per answer, with citations

Before
Answers about project status, decisions and owners were spread across Jira and Confluence.
My part
I built all 8 bases and the daily check alone.
Tools
Claude Code, Jira, Confluence, Git

Knowledge · B2B marketing services company

Daily AI news brief

One daily brief instead of dozens of sources.

  • Covered by 50 automated tests

How it works

  1. StartEvery day the brief collects news from 35 sources.
  2. AIA Claude Code agent groups the news by theme and links each item to live projects.
  3. ResultThe brief appears in Microsoft Teams.

35 sources → 1

brief a day

Before
Nobody had time to follow AI news or to see which news touched their projects.
My part
I built the whole system alone: talks with the team, design, build, testing, release and support.
Tools
Claude Code, Microsoft Teams

Knowledge · B2B marketing services company

Claude Code workshop

A hands-on workshop so a team can build with Claude Code on its own.

  • Exercises from real work

How it works

  1. PersonI prepared exercises from the team's real tasks.
  2. PersonI showed each step live.
  3. PersonI helped each person through the exercises.
  4. ResultI reviewed the errors with the group. The team now works with Claude Code on its own.

40 people

trained in one workshop

Before
The team did not use AI coding agents in daily work.
My part
I designed the exercises and ran the workshop alone.
Tools
Claude Code

Operations · AI automation agency

Broker update assistant

Every broker email about a load gets a fast and clear answer.

  • ~60 broker emails a day
  • No missed update requests

How it works

  1. StartA broker email arrives.
  2. AIAI finds update requests and load IDs. It pulls the driver, vehicle, route and status.
  3. ResultA structured message appears in the team chat.
  4. PersonA dispatcher replies in the chat.
  5. AIAI turns the reply into a professional email to the broker.
  6. RulesThe team gets an alert when a request waits too long.

1 more step in Details.

1 h → <5 min

to answer a broker

Before
Update requests got lost among hundreds of emails a day. Brokers waited up to an hour for an answer.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, LLM API, email, team chat

Sales · AI automation agency

Real-estate lead qualification

Each new buyer inquiry gets an answer and a qualification in minutes.

  • Lead conversion +30%

How it works

  1. StartA lead writes through the website, email or WhatsApp.
  2. AIAI asks about budget, location, property type and timeline, and suggests listings.
  3. RulesThe CRM gets the answers.
  4. PersonAn agent gets the qualified lead with a summary.

24-48 h → 5 min

to the first reply

Before
New leads waited a day or two for a first reply, and many went cold.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, LLM API, CRM, calendar

Operations · AI automation agency

Recruitment pipeline

AI screens résumés and books interviews.

  • Time to hire: 9-12 → 4-5 weeks

How it works

  1. StartApplications come in.
  2. AIAI ranks each applicant against the role.
  3. PersonThe hiring team approves the shortlist.
  4. ResultThe system books interviews in the calendar.

500 résumés

screened in 30 minutes

Before
Recruiters read each résumé by hand. Hiring took 9 to 12 weeks.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, LLM API, calendar

Operations · AI automation agency

Discovery with business owners

I find the step where AI pays off before anything is built.

  • Clients in 20+ industries
  • $245K a year saved and earned for 50+ client companies

How it works

  1. StartA business owner describes the problem.
  2. PersonI talk to the people who do the work and map the process as it is.
  3. RulesI size each step: time, volume and the cost of an error.
  4. ResultA short plan: what AI does, what rules do and where a person decides.

50+

discovery interviews with business owners

Before
Owners asked for AI without a clear process or a number to improve.
My part
I ran every interview and wrote each plan myself.
Tools
Interviews, process maps, n8n

Operations · AI automation agency

Junior team leadership

I led 3 junior specialists on client projects as Team Lead.

  • 3 junior specialists
  • 600+ h of hands-on mentoring
  • Weekly workflow reviews

How it works

  1. PersonI split client projects between the juniors by skill and load.
  2. PersonI worked next to each junior full time for 4 months.
  3. PersonEvery week I reviewed their workflows with them.
  4. ResultEvery project went live within 5% of the planned time.

12+ projects

in parallel, each within 5% of the planned time

Before
3 new junior specialists and 12+ client projects running at the same time.
My part
I led the team, split the projects and reviewed every workflow.
Tools
n8n, weekly reviews

Marketing · AI automation agency

Personal email campaigns

Weekly personal emails for each customer segment of an online store.

  • Open rate 12% → 25%
  • Email sales +15% in 3 months
  • Weekly instead of monthly

How it works

  1. StartData comes from the CRM and the online store.
  2. RulesCustomers are split into segments by purchases, browsing and location.
  3. AIAI writes product picks, subject lines and copy for each segment.
  4. ResultThe campaign goes out every week.

5 days → minutes

to prepare a newsletter

Before
One newsletter took 5 days to prepare, so the store sent one a month.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, CRM, e-commerce platform, LLM API

Marketing · AI automation agency

Telegram channel content

AI prepares posts for a Telegram channel. The owner only approves them.

  • Audience +14%
  • New channels: 0 → 10K subscribers in 2 months

How it works

  1. StartThe system collects posts from 15 reference channels.
  2. AIAI rewrites each post in the owner's style and keeps the image.
  3. AIAI scores each post for quality.
  4. PersonThe owner approves, edits or deletes each batch of 10 posts in Telegram.
  5. ResultApproved posts go to the publishing queue.

8 h → 30 min

a day to run the channel

Before
The owner spent 8 hours a day on ideas, visuals and posting.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, LLM API, Telegram

Marketing · AI automation agency

Content repurposing

One blog post or video becomes posts for each social platform.

  • 10+ h saved a week
  • Followers +50% in 3 months

How it works

  1. StartThe creator publishes a new blog post or video.
  2. AIAI finds the key ideas.
  3. AIAI writes an X thread, a LinkedIn post and an Instagram caption with hashtags.
  4. ResultThe posts are ready for each platform.

4x

more posts a week from one original

Before
The creator rewrote each piece by hand for each platform.
My part
I did it alone: discovery with the owner, design, build, testing and handover to the client's team.
Tools
n8n, LLM API, social media APIs

Operations · AI automation agency

Data migration to Airtable

Records from old systems moved into Airtable without stopping work.

  • Fixed the OAuth and rate-limit stops behind 12+ h of downtime a month
  • 100+ APIs and CRMs connected

How it works

  1. StartRecords come from old systems and the Super Dispatch TMS.
  2. RulesThe sync handles OAuth and API rate limits, so it does not stop.
  3. ResultAll records are in Airtable and stay in sync.

100K records

moved to Airtable

Before
Syncs stopped on OAuth and rate-limit errors, with 12+ hours of downtime a month.
My part
I found the cause of the stops and fixed the OAuth and rate-limit handling alone.
Tools
n8n, Airtable, REST APIs, OAuth 2.0

My own app

AiMax Eye: the MacBook notch as a work tool

A native macOS app. On hover, the notch opens a panel with a player, a file shelf, clipboard history, snippets, the next meetings and translation.

  • Native SwiftUI and AppKit
  • Join a Zoom, Meet or Teams call in one click
  • AI translation with Claude through n8n

How it works

  1. StartThe pointer reaches the notch.
  2. RulesThe panel opens. Tabs switch on hover: music, shelf, clipboard, snippets, calendar, translate, notes.
  3. RulesOffline translation runs on macOS itself, English to Russian and back.
  4. AIOn a button, a second mode sends the text to Claude with a personal prompt through an n8n flow with Supabase.
  5. ResultAutomated tests cover the snippets format, translation routing, meeting links and the countdown.

0.0% CPU

at rest, ~40 MB of memory, 2.1 MB app

Before
Files, clipboard, meeting links and translation lived in different apps and windows.
My part
My own project. I designed it and built it with Claude Code: the native app and its n8n backend.
Tools
Swift 6, SwiftUI, AppKit, Claude Code, n8n, Claude Haiku, Supabase