Aleksejs Buss — Agentic AI Engineer · Multi-Agent Orchestration

§ I — At a glance

01/06

A short ledger of what shipped.

Every figure points to live infrastructure — production-grade systems I engineered, built, and deployed end-to-end under my own name.

01.

5

Production systems

All deployed & live solo

02.

5400+

Automated tests

Vitest · Playwright · Pytest

03.

10mo

Zero to five live

Solo delivery, since 07/2025

04.

4

Languages spoken

RU · LV · EN · DE

§ II — The Way

02/06

Fourteen years in logistics.
Now I build AI systems for production.

ALEKSEJS BUSS · AGENTIC AI ENGINEER

I didn't take the conventional route. I took the self-disciplined one.

Fourteen years in multicultural logistics — Tesco Carlisle, long-haul freight across the UK and Europe — taught me systems, processes, and how to deliver under pressure.

In 2022 I retrained as a Frontend Developer at Tel-Ran Berlin (960h, DEKRA-certified). Then I went deeper: LLMs, prompt engineering, serverless, local AI. Nights and weekends, while driving part-time to pay the bills.

From mid-2025, I've been shipping. Not learning. Real systems, real infrastructure, built end-to-end — under my own name.

“Self-discipline is the skill that transfers. Everything else I learn fast — and I have.”

2008 — 22

Logistics, UK & Germany

Tesco Carlisle (WMS, planograms), long-haul freight. Process, resilience, multicultural teams.

10/22 — 10/23

Frontend Developer · Tel-Ran Berlin

DEKRA-certified, 960h. JS Advanced, React, DSA, DevOps, QA, UI/UX.

2023 — 25

AI system design & engineering self-study

n8n, Cloudflare Workers, Ollama, prompt engineering, local LLM, Docker.

07/25 — now

Engineering & operations · 5 production systems

End-to-end: system design → deployment → monitoring. Designed and operated solo.

Now

Open to your team

Immediate availability. Germany-based, EU citizen. Remote or on-site.

§ III — Approach

03/06

What I build
— and why it works.

I design AI systems from first principles — agent topology, provider failover, latency budgets, cost routing — then build and operate them end-to-end. Six things I bring on day one.

01. Ship rate

Founder-grade ownership.

Spec → architecture → code → deploy → monitor → iterate. No hand-offs, no "not my job". Five systems shipped solo since mid-2025. The cadence your runway needs.

02. Full stack

AI, backend & infra in one.

TypeScript/Next.js, LLM pipelines, multi-agent orchestration, FastAPI/Python, Cloudflare edge, Docker/systemd. One headcount, full stack — your seed runway stretches further.

03. AI-native

LLM is native, not a bolt-on.

I build agent systems, not wrappers: Mixture-of-Agents pipelines, disagreement detection, reflection loops. Multi-provider failover (OpenAI / Gemini / Groq / Ollama), structured outputs, voice pipelines. Designed with latency, cost & failure modes in mind.

04. Production hygiene

Defaults, not afterthoughts.

HMAC-SHA256, zero-secrets architecture, rate limiting, systemd auto-restart, health checks, webhook verification, secrets via env. Built in from day one.

05. Late-career grit

Self-driven, under pressure.

Career-switched at 35+ and shipped five systems solo while moving from part-time driving into full-time engineering. Not talent — discipline. I don't need reminders to unblock myself.

06. Full product

Beyond the backend.

I can design, implement, and launch an AI product end to end — voice, visual generation (ComfyUI, Stable Diffusion), UI, copy, deployment. Useful when your team is three people and a deadline.

§ IV — Work · Selected

04/06

Production systems,
shipped solo.

A selection from five production systems below — each deployed on infrastructure I own and operate, engineered and built end-to-end to production standards.

01 05 / 2026 — ongoing Open Source · Flagship

Orchestra — Multi-Agent AI Workspace

Mixture-of-Agents pipeline · code-guaranteed Skeptic · live cost telemetry.

Self-hosted AI workspace that runs 3–5 specialised expert agents in parallel on every substantive turn. Embedding-based disagreement detection surfaces expert conflict instead of smoothing it away; a reflection critic revises the final answer. 4,100+ tests, 70+ documented post-mortems, CI — engineering-led, MIT-licensed.

Router (DPG)→proposers ∥ Skeptic→disagreement detector→Aggregator→reflection

Read case study →

live

4100+tests

70+post-mortems

5parallel agents

MITopen source

TypeScript Next.js 15 MoA Vitest Playwright

02 08 / 2025 — ongoing Open Source · Flagship

4take — Multi-Model AI Code Review Council

FastMCP & CLI council · 4-model consensus protocol · adversarial synthesis.

Autonomous multi-model AI code review council querying 4 independent LLMs in parallel. Scores confidence, discovers consensus & critical regressions before code merges. Zero-dependency FastMCP server + high-speed CLI, comprehensive Pytest suite, MIT-licensed.

CLI / FastMCP→Parallel LLM Pool (4 models)→Tokenizer→Consensus Synthesizer→Review Matrix

Read case study →

live

Pytestsuite

4parallel LLMs

FastMCPprotocol

MITopen source

Python FastMCP Pydantic v2 Pytest

03 07 / 2025 — ongoing Production SaaS

AI Dictaphone v7.0 — Telegram Mini-App Voice Agent

Voice notes to structured intelligence via Groq Whisper & Gemini 2.5.

Mobile-first voice transcription and structured extraction app embedded into Telegram. Streams audio directly to Groq Whisper with fallback to OpenAI Whisper, then runs Gemini 2.5 structured output pipelines with HMAC-SHA256 authenticated webhook sync.

TMA Client (WAV)→Cloudflare Worker→Groq Whisper (failover OpenAI)→Gemini 2.5 (JSON Schema)→Telegram Bot

Read case study →

live

<800msSTT latency

99.9%uptime

Dualprovider STT

LiveSaaS

TypeScript CF Workers Whisper Gemini 2.5

04 09 / 2025 — ongoing Production Bot

AI Moderator Bot — Context-Aware Community Guardian

LLM moderation pipeline with n8n orchestrator & multi-turn context memory.

Autonomous Telegram community moderation bot operating on zero-downtime n8n workflows with LLaMA 3.3 and Gemini fallbacks. Detects spam, scams, toxic intent, and multi-turn manipulation patterns in real-time. 1,000+ tests across bot, API, and mini-app.

Telegram Webhook→n8n Orchestrator→LLaMA 3.3 70B (failover Gemini)→Decision Action

Read case study →

live

1000+tests

24/7active

n8norchestrator

Multi-turnmemory

n8n LLaMA 3.3 Docker

05 11 / 2025 — ongoing Commercial SaaS

AI Psychology Bot — Empathetic Micro-Counselor

Therapeutic reflection loops with Telegram Stars native monetization.

Commercial AI psychological assistant featuring safe reflective listening techniques, personalized state tracking, and integrated Telegram Stars payments for subscription tiers.

User Session→Safety Classifier→Empathetic Reasoning Loop→Telegram Stars Billing

Read case study →

live

Starsmonetized

100%private

Safetyguardrails

LiveSaaS

TypeScript Telegram Stars Firebase

§ V — Toolkit & Stack

05/06

Production stack
— tools I ship with.

Tools I've used in production — tested, debugged under pressure, and operated on live infrastructure.

01.

AI · LLM · Agents

Mixture-of-Agents (MoA) FastMCP Protocol Disagreement Detection OpenAI GPT-4o Google Gemini 2.5 Pro Groq Whisper Ollama Local LLM RAG Embeddings

02.

Programming & Testing

TypeScript (Strict) Python 3.12+ (Pydantic v2) Next.js 15 (App Router) React 19 FastAPI Tailwind CSS / Vanilla CSS Three.js

03.

Automation & Workflows

Cloudflare Workers n8n Self-Hosted Telegram Bot API Telegram Mini-Apps Supabase Firebase RTDB

04.

Security & Protocols

Vitest (Unit) Playwright (E2E) Pytest HMAC-SHA256 Auth Docker & Compose systemd Services

RU

Russian

Native fluency

LV

Latvian

Native fluency

EN

English

C1 Professional working proficiency

DE

German

B1 Independent user

Autonomous AI systems, engineered for production.

A short ledger of what shipped.

Fourteen years in logistics.
Now I build AI systems for production.

What I build
— and why it works.

Founder-grade ownership.

AI, backend & infra in one.

LLM is native, not a bolt-on.

Defaults, not afterthoughts.

Self-driven, under pressure.

Beyond the backend.

Production systems,
shipped solo.

Orchestra — Multi-Agent AI Workspace

4take — Multi-Model AI Code Review Council

AI Dictaphone v7.0 — Telegram Mini-App Voice Agent

AI Moderator Bot — Context-Aware Community Guardian

AI Psychology Bot — Empathetic Micro-Counselor

Production stack
— tools I ship with.

Let's build something impossible.

Autonomous AI systems, engineered for production.

A short ledger of what shipped.

Fourteen years in logistics.Now I build AI systems for production.

What I build— and why it works.

Founder-grade ownership.

AI, backend & infra in one.

LLM is native, not a bolt-on.

Defaults, not afterthoughts.

Self-driven, under pressure.

Beyond the backend.

Production systems,shipped solo.

Orchestra — Multi-Agent AI Workspace

4take — Multi-Model AI Code Review Council

AI Dictaphone v7.0 — Telegram Mini-App Voice Agent

AI Moderator Bot — Context-Aware Community Guardian

AI Psychology Bot — Empathetic Micro-Counselor

Production stack— tools I ship with.

Let's build something impossible.

Fourteen years in logistics.
Now I build AI systems for production.

What I build
— and why it works.

Production systems,
shipped solo.

Production stack
— tools I ship with.