Curriculum Vitae · № 2026 / CV-01 · Updated June 2026 · Deutsche Version →

Aleksejs Buss

AI Engineer · Multi-Agent Systems

Creator of Orchestra — open-source Mixture-of-Agents platform.
Eight production systems. Shipped solo.

Location Hof / Saale, DE · EU citizen
Status Available immediately · remote or on-site
§ 01

Summary — at a glance

AI engineer specialising in multi-agent LLM systems. Creator of Orchestra — an open-source Mixture-of-Agents workspace (TypeScript / Next.js 15) with 2,600+ tests and 70+ documented post-mortems: parallel expert agents, a code-guaranteed Skeptic, embedding-based disagreement detection. Since mid-2025 I have shipped eight production-grade AI systems — built and deployed end-to-end — Cloudflare Workers edge, Telegram Mini-Apps, multi-provider LLM pipelines, payments, security. Features from spec to live infrastructure, end to end — founder-grade ownership at one headcount.

§ 02

Selected work — production systems

05/2026 — now Flagship · OSS

Orchestra — Multi-Agent AI Workspace

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

Self-hosted AI workspace running 3–5 specialised expert agents in parallel on every substantive turn. Router with dynamic persona generation; a Skeptic guaranteed by code, not by prompt; embedding-based disagreement detection that surfaces expert conflict instead of smoothing it away; reflection critic + revisor pass. Live per-chat token + USD cost banner. BYOK or fully local via Ollama. 2,600+ tests, 70+ documented post-mortems, CI — MIT-licensed.

Router (DPG)proposers ∥ Skepticdisagreement detectorAggregatorreflection

08/2025 — now Production

AI Dictaphone v7.0

Edge voice pipeline · Whisper → LLM · sub-second latency.

Serverless voice-to-insight pipeline on Cloudflare edge. Voice note → Groq Whisper transcription → LLaMA 3.3 70B summarization → rich Mini-App UI. Zero cold-start, HMAC-verified webhooks, no persistent storage of voice content.

Telegram Mini-AppCloudflare WorkersGroq Whisper+LLaMA 3.3 70B

09/2025 — now

AI Moderation Bot

n8n workflow · dual-AI failover · auto-escalation.

15+ node n8n pipeline classifies every Telegram message with structured JSON outputs, routes to action, escalates sanctions across three tiers. Dual-provider failover (Nemotron primary, GPT-4.1 fallback) keeps moderation alive when any single LLM API is down.

MessageNemotronfallback GPT-4.1Routeraction

10/2025 — now SaaS product

AI Psychology Bot

Full SaaS · payments · crisis support · CBT exercises.

End-to-end SaaS solo: HMAC-SHA256 auth → distributed Apps Script backend → Firebase Realtime DB → Gemini 2.5 + TTS pipeline. Telegram Stars for native checkout, atomic credit operations, referral program, panic intervention, 7-Skill protocol.

Mini-AppHMAC-SHA256Apps ScriptFirebaseGemini+TTS

02/2025 — now bushmark.cc

Bushmark — Secure Clipboard

Self-hosted · cookie auth · one-click install.

Self-hosted secure clipboard at bushmark.cc — three persistent slots, 24h history, auto-rotation, Tailwind UI. One-line bash install on any Ubuntu VPS handles Nginx + FastAPI + systemd setup automatically.

BrowserNginx reverse proxyFastAPI (systemd)

07/2025 — now Self-hosted

CyberDed Ultra + ComfyUI

Local AI infra · one-click installer · Android + Linux.

Bash installer for the entire local-AI stack: Ollama for inference, ComfyUI for image generation, NVIDIA Container Toolkit auto-config, Flask backend with Server-Sent Events streaming chat, real-time monitoring. Two OS targets — Linux + Termux Android.

Bash installerOllama+ComfyUI DockerFlask SSE

11/2025 — now Open Source

Termux AI Dictaphone

Android as offline AI server · zero cloud cost.

Complete offline AI on any modern Android: Whisper transcription via faster-whisper / whisper.cpp, LLaMA / Gemma inference via Ollama, all on-device through Termux. Zero API cost, full privacy. Published OSS.

Termux (Android)faster-whisperOllama (LLaMA, Gemma)

§ 03

Experience — the route

07/2025 — now

AI Engineer · Solo Production

Independent (full-time since 01/2026) · Hof / Saale, DE
  • Designed, deployed, and operate eight production AI systems end-to-end — own architecture, own code, own infrastructure.
  • Built Orchestra, an open-source multi-agent AI workspace (TypeScript/Next.js 15): MoA pipeline with parallel expert proposers, code-guaranteed Skeptic, embedding-based disagreement detection — 2,600+ tests, 70+ post-mortems, CI.
  • Multi-provider LLM pipelines (OpenAI, Gemini, Groq, Ollama) with failover; structured outputs; voice + vision flows.
  • Edge serverless on Cloudflare Workers; Telegram Mini-App + Bot integration with HMAC-SHA256 webhook verification.
  • Self-hosted infra: Ubuntu VPS, Nginx, systemd, Docker, Tailscale, NVIDIA Container Toolkit.
  • Monetization: Telegram Stars payments, atomic credit ops, referral logic, anti-fraud heuristics.
11/2023 — 12/2025

Professional Driver · part-time (alongside AI upskilling)

L.O.-Transport GmbH · Tauperlitz, DE
  • Part-time (50%) to fund the transition into AI engineering — remaining time invested consistently in building production AI systems.
  • In parallel, self-directed: n8n agent workflows, local LLM deployment (Ollama, llama.cpp, faster-whisper), Cloudflare Workers, FastAPI, edge-first architectures.
  • Full-time as an independent AI Engineer since 01/2026 (see above).
10/2022 — 10/2023

Frontend Developer (retraining)

Tel-Ran Berlin · 960h · DEKRA-certified
  • JavaScript Advanced (ESM, async, prototypes); React fundamentals; DSA; HTML5/CSS3.
  • DevOps essentials (Git/GitHub, CI fundamentals, Docker basics); QA; UI/UX foundations.
  • Career switch from logistics into software engineering.
2008 — 2022

Logistics & Long-Haul Freight

Tesco Carlisle (UK) · BRS UK · Deutsche Post (subcontractor) et al.
  • Warehouse logistics & systems merchandising at Tesco (2008–2016): warehouse management systems (WMS), planograms.
  • Long-haul freight thereafter (2016–2022) across multicultural teams in UK, DE, NL, FR, BE.
  • Process discipline and resilience under deadline pressure — the skill that transfers.
§ 04

Stack — used in production

Agent Systems
Mixture-of-Agents (MoA) orchestration · dynamic persona generation · embedding-based disagreement detection · reflection loops (Generator-Critic-Revisor) · cost telemetry · agent tool design · sandbox & SSRF guards
AI / LLM
OpenAI GPT-4 / 4.1 / TTS · Google Gemini 2.5 · Groq Whisper-large-v3 · Ollama (LLaMA, Gemma) · faster-whisper · whisper.cpp · OpenRouter · Nemotron · multi-provider failover · structured outputs · prompt engineering · RAG-ready patterns
Automation
n8n · workflows · agents · code nodes (JS/Python) · webhook routing · error handling & retries · chain-of-thought architecture · multi-protocol agents
Languages
TypeScript (strict) · Next.js 15 / React · JavaScript (ESM, async) · Python (FastAPI, Flask) · Bash · HTML5 · CSS3 · SQL · streaming APIs (SSE)
Cloud / Edge
Cloudflare Workers · Wrangler · Google Apps Script · Firebase Realtime DB · GitHub Pages · Supabase / PostgreSQL
Operations
Docker / Compose · Nginx (reverse proxy, TLS) · Ubuntu VPS · systemd (services, timers) · health checks · auto-restart · log rotation · NVIDIA Container Toolkit · Tailscale
Security
HMAC-SHA256 · JWT · Telegram Bot API · Telegram Mini-Apps SDK · Telegram Stars payments · CSP · CORS · webhook verification · rate limiting · secret management · anti-fraud logic
Tooling
Git / GitHub · GitHub Actions CI · Vite · Vitest · Playwright · ESLint · Wrangler · macOS / Linux daily driver
§ 05

Education & languages

Frontend Developer

10/2022 — 10/2023 · 960 hours

Tel-Ran Berlin · DEKRA-certified vocational programme. JavaScript Advanced, React, DSA, DevOps, QA, UI/UX, HTML5/CSS3.

Certification: DEKRA № 2023337
Issued: Berlin, Germany — 2023

Languages

Russian Native
Latvian C2 · Fluent
English C1 · Business
Deutsch B1