maxtrifonov.dev
DE
Email me
Portrait of Max Trifonov

Max Trifonov, M.Sc. Full-Stack AI Engineer LLM integration and automation Based in Düsseldorf, Germany, working remotely

I build AI that makes it into production. And keeps running there.

Plenty of companies have introduced AI, few use it day to day. I automate recurring workflows and bring AI into the systems you already use, up to reviewing customer conversations, and I keep everything running reliably.

Email me View projects
  • 20,000+ automated conversation analyses
  • 2,000+ new transcripts every month
  • 17 live services and systems built
  • Voice AI for real-time conversations

Why it pays off

Time

Follow-up, notes and moving data between systems happen automatically. People stay focused on the conversation.

Skills shortage

Unfilled positions cost the German economy around 49 billion euros in lost output in 2024.1 Automation takes load off the team you already have.

Fewer errors

Data moves between systems without retyping, and every step is logged.

AI that actually runs

57 percent of German companies use AI, yet only 3 percent say they are making strong use of its potential.2

When it is not worth it

  • Rare tasks: anything that only comes up a few times a month is cheaper by hand.
  • No fixed process: without clear steps, there is nothing that can be reliably automated or measured. Process first, then AI.
  • Decisions about people: a person makes the final call. AI does not take responsibility off anyone's shoulders.

1 German Economic Institute (IW), press release of 12 May 2024. 2 Bitkom, survey of 603 companies with 20 or more employees, press release of 14 September 2026.

How I can help

Example run
  1. 09:41:07Email receivedwith attachment
  2. 09:41:08Attachment recognizedorder, 2 pages
  3. 09:41:10Data transferredinto the customer system
  4. 09:41:12Document createdorder confirmation
  5. 09:41:121 questionfor the team
  6. 09:41:13doneafter 6 seconds

For companies in any industry where work repeats itself: automating workflows, bringing AI into existing systems and, where customer conversations happen, reviewing those too.

Workflow automation

Generate documents, sort emails and move data between systems, with n8n or custom code.

AI in your existing systems

API integrations, production monitoring and hosting in the EU.

Conversation analytics

Automatically review sales, advisory or service calls, with a clear rationale for every result.

Voice AI and training

Simulate conversations in real time, so your team can practice without real customers.

Projects

What each system does and the stack behind it.

  1. 20,000+

    analyses, 2,000+ transcripts a month

    AI conversation analysis

    Automatically reviews advisory calls against the conversation guide: what was covered and where things went wrong. Team leads see the results for their team, for every call instead of a few spot checks.

    Stack Gemini, Claude, Node.js, Supabase, Google Cloud Run

    Read the case study

  2. Automated

    documents, forms, workflows

    Automation

    Generates documents from customer data, fills in forms, applications and calculators, and runs recurring workflows in the background, saving hours of routine work.

    Stack Python, Selenium, n8n

  3. Real-time

    voice conversations with AI customers

    AI call training assistant

    Advisors practice phone calls with AI customers who respond like real people, and get feedback after every call. New advisors practice before they talk to real customers.

    Stack Gemini Live, Vertex AI, React, WebSockets, Supabase, Google Cloud Run

  4. Live

    guidance during the call

    Real-time coach for advisory calls

    Supports advisors during live calls with prompts while the conversation is still happening.

    Stack Electron, Gemini Live

  5. 17

    systems and services

    Systems and services in operation

    Web apps, automations, bots and desktop programs, including an app platform with single sign-on that brings a company's internal apps together, with roles and permissions for each team.

    Stack Google Cloud Run, Cloud Build, Supabase, Node.js, React, Electron, Python, n8n, Telegram, Docker, Hetzner, Gemini, Claude

  6. Documents

    extracted and evaluated

    AI document service

    Extracts information from documents and produces an evaluation that a person reviews before it is used.

    Stack Python, FastAPI

Confidentiality and privacy

Confidential

Client projects, documents, data and agreements stay confidential, which is why you won't find any names here. Every engagement starts with a non-disclosure agreement.

Privacy by design

Customer and employee data is processed only as far as the purpose requires. Access on a need-to-know basis, EU hosting wherever possible, lean logs with short retention, and a data processing agreement with every provider.

Built to the EU AI Act

Every system I build follows the EU AI Act. Risk classification and obligations are settled before development starts, with transparency for everyone involved and human oversight. When AI evaluates conversations or employees, a person makes the final call.

Fair to the people being assessed

When AI assesses people, every result has to be explainable and verifiable. Safeguards against unfair or incorrect results are built into the system.

How I work

Documentation is part of the deliverable

Architecture, decisions, runbooks and rollback procedures are written down and kept current. Whoever takes over understands what was built and why.

Traceable

Every AI result comes with its reasoning and evidence. Every change is logged and can be traced back.

Thorough

Measure first, then make claims. I come from biomedical research, so every claim about an AI system needs data behind it, and every major change is independently reviewed and tested before release.

Operations included

Monitoring, alerting, and a way back when something goes wrong.

Clear recommendations, not a list of options

I tell you what I would do, and why.

Both sides of the table

I have worked as an advisor myself, so I know the conversations my systems analyze.

Background

  1. 2023 M.Sc. in Biology, Heinrich Heine University DüsseldorfPython, R, statistical data analysis
  2. 2023 to 2024 Diabetes research
  3. Since 2024 Moved into advisory work
  4. Since 2025 Building AI systems for businesses
  5. Since 2026 Freelance full-stack AI engineer

Tech stack

AI and language models

Claude, Gemini, Gemini Live, Vertex AI (EU), Whisper

Multi-stage analysis with independent cross-checks, real-time voice AI, tests that guard against incorrect assessments, and cost control through model selection and prompt caching.

Backend and APIs

Node.js, Express, Python, FastAPI, Flask

REST APIs, webhooks, WebSockets and OAuth integrations with services like Zoom.

Frontend

React, Tailwind CSS, Electron

Web and desktop apps with login, roles and tailored views for different user groups.

Data

Supabase, PostgreSQL

Data modeling, row-level security and scheduled jobs that run directly in the database.

Cloud and operations

Google Cloud Run, Cloud Build, Docker, Hetzner, Linux

Automated deployments, monitoring and alerting, health checks, instant rollback and load testing before launch.

Automation

n8n, Playwright, Selenium

Workflows across systems, plus browser automation for apps that have no API.

Engineering practice

Git, GitHub, Jira, Claude Code

Feature branches, pull request reviews and AI-assisted development.

Get in touch

Need a system that runs reliably? A short email is all it takes.

Email me
WhatsApp
+49 1575 0752500
Location
Düsseldorf, Germany. Remote projects welcome.