Menasse Gebregzi
Get in Touch

Cologne, Germany — Open to remote & on-site engagements

Cloud & AI
Solutions Builder

Building intelligent systems with AWS, Generative AI, Python and FastAPI. Turning ideas into production-ready cloud architecture.

AWSCertified (in progress)
Bedrock& GenAI
Python& FastAPI
CloudArchitecture
Menasse Gebregzi — Cloud & AI Solutions Builder, Cologne

Menasse Gebregzi — Cologne, Germany

About

Full-stack roots, cloud and AI in progress.

Menasse Gebregzi is a full-stack developer based in Cologne, Germany, transitioning into Cloud & AI engineering. He came to infrastructure the long way around — twenty years in media design and full-stack development, shipping under real deadlines to real audiences. That order matters more than it looks: he learned what users do to a system long before he learned how to diagram one.

Today he builds and operates production systems end to end — most recently Atrium CRM, a multi-tenant SaaS for German real estate agents built with FastAPI, PostgreSQL, Docker, and a full monitoring stack. Designed, built, migrated, and operated solo: from data model to deployment to infrastructure ownership. Being the developer and the person on call at the same time changes which decisions look right.

Currently preparing for the AWS Certified Generative AI Developer – Professional certification, building daily with the AWS CLI, boto3, and IAM. Theory goes straight into something that runs.

20+

Years shipping under real deadlines

Approach

Architecture First

Every system starts with a clear design. Infrastructure decisions are made deliberately, not by accident.

Ship & Learn

Production is the best teacher. Build fast, deploy early, iterate based on real feedback.

Full Stack Thinking

From Bedrock API to React frontend — understanding the whole system leads to better decisions at every layer.

Documentation

Code that isn't explained doesn't scale. Every project includes architecture diagrams and clear reasoning.

Approach

More than code.
Understanding.
Building.
Shipping.

Cloud and AI engineering is more than writing functions and deploying containers. It is about understanding what a system needs to do, why it needs to do it that way, and how every architectural decision affects reliability, cost, and maintainability. Every project starts with a clear question: what problem does this solve, and what is the simplest architecture that solves it well?

“The best architecture is the one you can explain to someone else.”

Projects

Built. Shipped. Learned.

Project 01 — Production SaaS · Live

Live

Atrium CRM — Multi-Tenant Lead Management for Real Estate

atrium-crm.de ↗

Full-stack SaaS CRM built for German real estate agents. Multi-tenant architecture with row-level security on all 23 database tables, Stripe subscription billing, and a monitoring stack in production. Designed, built, operated, and migrated end to end — from data model to deployment to infrastructure ownership.

Python · FastAPI · Next.js · PostgreSQL · Redis · Docker · Nginx · Stripe · Prometheus · Grafana · GitHub Actions · dotenvx · IONOS VPS

Architecture

Multi-tenant FastAPI backend → PostgreSQL with RLS (tenant_isolation policy on all 23 tables) + Redis rate limiting → Next.js frontend → Nginx reverse proxy. Stripe webhooks handle the full subscription lifecycle. Monitoring via Prometheus + Grafana + Node/Postgres/Redis exporters. Automated daily database backups to Hetzner Object Storage (S3).

Infrastructure Migration

The backend originally ran on Railway. Migrated to a self-managed VPS for full control over infrastructure, costs, and data residency. Parallel run: both environments operated side by side before cutover. Secrets migrated without disruption. Zero-downtime switch via DNS. Legacy decommissioned: stale DNS cleaned, credentials rotated. Result: zero downtime · ~83% lower hosting costs · no vendor lock-in.

What was learned

Why RLS at database level is safer than application-layer filtering: no code path can accidentally leak cross-tenant data. How Stripe's async webhook model requires idempotent handlers. Why docker restart doesn't re-read --env-file. That a migration isn't finished at cutover — decommissioning (stale DNS, old credentials) is part of the job, and skipping it leaves a security gap.

Design Decision

Infrastructure managed as code in a dedicated repo (atrium-infra): docker-compose, nginx config, deploy scripts, and encrypted secrets (dotenvx) — all versioned and deployed via GitHub Actions on every push to main. PostgreSQL RLS chosen over middleware-based tenant filtering: enforcement at DB level means no code path can leak cross-tenant data. Self-managed hosting over PaaS: lower cost and full infrastructure ownership in exchange for owning backups, patching, and monitoring.

Project 02 — 2026

Complete

KitaDoc — Documentation Platform for Early Childhood Education

kita.eyedia.de ↗

A modern web application for daycare centres that digitises the pedagogical day-to-day. KitaDoc enables structured documentation of child observations, developmental progress, and key care information — with a clear interface that reduces administrative overhead and keeps educators focused on what matters.

Next.js · TypeScript · Supabase · PostgreSQL · Row-Level Security · PM2 · Node.js · VPS

Architecture

Next.js App Router → Supabase (Auth + PostgreSQL with RLS) → PM2 process manager on VPS. Child data is isolated per organisation via row-level security policies. Development tracking uses a multi-criteria scoring model aligned to the NRW early education framework.

What was learned

How Supabase client recreation on every render causes infinite useEffect loops — fixed with useMemo. Why DELETE + INSERT is safer than upsert when no unique constraint exists. How to handle coexisting data ID schemes during a live migration without breaking existing records.

Design Decision

Supabase chosen for built-in auth, RLS, and real-time in a single managed service. For a product handling child data under GDPR, row-level security enforced at database level is a non-negotiable baseline — not an afterthought.

Project 03 — Foundation

Complete

Bedrock Chatbot — Single User

First direct integration with Amazon Bedrock. A command-line Python chatbot using boto3 and Nova Micro with persistent conversation history. The foundation for understanding how LLM APIs work and how context windows handle memory.

Python · boto3 · Amazon Bedrock · Nova Micro · eu-central-1

Architecture

Single process → boto3 client → Bedrock Runtime → Nova Micro. No server, no sessions, no concurrency.

What was learned

How the AWS SDK authenticates, how inference profiles work, why conversation history must be sent with every request, and how temperature affects output quality.

Project 04 — Production Step

Complete

Bedrock Chatbot API — Multi User

Evolved the single-user chatbot into a production-ready REST API using FastAPI and uvicorn. Multiple users can maintain independent conversations simultaneously via session IDs. Exposed as HTTP endpoints with automatic OpenAPI documentation.

Python · FastAPI · uvicorn · boto3 · Amazon Bedrock · REST API · Session Management

Architecture

HTTP POST /chat → FastAPI router → session store (in-memory dict) → boto3 → Bedrock Runtime → JSON response. Each session_id maps to its own conversation_history list.

What was learned

How to wrap an LLM call in a web framework, how session isolation works, why stateless HTTP needs explicit state management, and how FastAPI's automatic /docs endpoint accelerates testing.

Design Decision

Chose FastAPI over Flask for async support, Pydantic validation, and automatic OpenAPI docs — all relevant for a production AI API.

More projects in development. Each one builds on the last.

Curriculum Vitae

Menasse Gebregzi

Cloud & AI Solutions Builder — Cologne, Germany

Contact

LocationCologne, Germany
BornFebruary 24, 1973 — Asmara, Eritrea
NationalityGerman, Eritrean

Languages

GermanNative
EnglishProfessional
TigrinyaNative

Education & Certifications

AWS Solutions Architect – Professional

Planned · 2027

AWS Certified DevOps Engineer – Professional

Planned · 2027

AWS Generative AI Developer – Professional

In Progress · 2026

Digital Film & Animation (Diploma)

SAE Institute Cologne, 2010 – 2011

Media Design — IHK Certified (Chamber of Commerce)

Macromedia Academy Stuttgart, 2000 – 2002

Photographer Training

Johannes Gutenbergschule & Axel Waldecker Fotografie Stuttgart, Ludwigsburg, 1994 – 1996

Cloud Solutions

Amazon BedrockEC2 / VPC / IAMS3CloudWatchLambdaAWS CLITerraform

AI Solutions

Prompt EngineeringRAG ArchitectureAmazon NovaMulti-turn ChatSession MemoryLLM APIs

Backend & Infrastructure

PythonFastAPIREST APIsDockerCI / CDPostgreSQL / SQLiteCaddy

Frontend & Full Stack

Next.js / ReactTypeScriptNode.jsTailwind CSSVercel / VPS Deploy

Selected Projects

eyedia.de — CMS & Blog platform (Next.js, SQLite, Docker)
Atrium CRM — Lead management SaaS (FastAPI, PostgreSQL, Stripe)
Bedrock Chatbot API — Multi-user GenAI API (Python, FastAPI, Amazon Bedrock)
KitaDoc — Documentation platform (Next.js, Supabase)

Stack

Tools built for production.

01

AWS Bedrock

Foundation Models as API. Integrating Claude, Nova, and Llama into production systems via boto3 and REST.

02

Python & FastAPI

Backend services that are fast, typed, and documented. REST APIs with async support and automatic OpenAPI docs.

03

Cloud Architecture

VPC, EC2, S3, IAM. Designing secure, scalable infrastructure on AWS with least-privilege access.

04

Generative AI

Prompt engineering, RAG systems, multi-user chat with session memory. Building AI that works in the real world.

05

Docker & Deploy

Containerized applications deployed via git-based workflows. Caddy reverse proxy, automated builds.

06

Next.js & React

Full-stack TypeScript applications. From API routes to responsive frontend — end to end.

07

System Design

Architecture diagrams, sequence diagrams, data flow. Communicating technical decisions clearly.

08

Always Learning

AWS certifications in progress. Daily hands-on practice. The AWS ecosystem is vast — curiosity drives the exploration.

AWS. Python. FastAPI. Bedrock. Production.

Experience

Engineering and building.

2026 – Present

Cloud & AI Solutions Builder (Independent)

Building cloud-native applications and AI integrations. AWS Bedrock, FastAPI, Python. Pursuing AWS certifications.

2020 – Present

Full-Stack Developer & Digital Product Builder

Designed and built multiple SaaS products including eyedia.de (Next.js CMS), Atrium CRM (FastAPI + PostgreSQL), KitaDoc (Next.js). Deployed on Hetzner VPS with Docker and Caddy.

2005 – 2015

Film & Media Production

Documentary filmmaking and commercial video. Directed content for corporate clients. Built strong visual communication and storytelling skills.

2002 – 2004

Media Designer — Schönstein Kommunikationsdesign, Cologne

Interactive media development. Flash developer and Co-Director on industrial commercial film.

Certifications

AWS Certification Path

A focused path through the AWS certification ecosystem — from Generative AI to cloud architecture and DevOps.

In Progress · 2026

AWS Generative AI Developer – Professional

Building production-grade GenAI applications on AWS. Amazon Bedrock, prompt engineering, RAG architectures, and responsible AI.

Planned · 2027

AWS Solutions Architect – Professional

Designing resilient, cost-efficient architectures. VPC, EC2, RDS, CloudFront, IAM, and high-availability patterns.

Planned · 2027

AWS Certified DevOps Engineer – Professional

Continuous delivery, automation, and monitoring on AWS. CodePipeline, CloudFormation, ECS, and operational excellence.

All certifications are pursued through hands-on project work, not just study materials.

Contact

Available for cloud architecture and AI engineering engagements. Based in Cologne, working across the DACH region — remote or on-site.

Emailkontakt@eyedia.de
Phone+49 178 82 50 869
LocationCologne, Germany
AvailableDACH region — remote & on-site

I respond to all inquiries within 24 hours. For urgent assignments, feel free to call directly.