Shipped Projects

Cloud services, control workflows, operators, multi-tenant systems and the tooling around them. Each one shipped to production.

View Projects Start a Project

Projects

Cloud services, control workflows and the systems around them - designed and shipped to production. Code, docs and runbooks included.

AI Copilot

Contenox — Open-Source AI Agent Runtime

TL;DR: Local-first AI agent runtime that compiles to a single Go binary and configures behaviors via versioned YAML configuration files instead of proprietary code.

The Problem: Deploying and scaling AI agents usually requires complex Python/Node runtimes, vendor-locked API gateways, and heavyweight infrastructure, leading to high latency and deployment overhead.
The Implementation: Architected a modular Go engine with policy-as-config execution chains. Integrated direct LLM client interfaces (llama.cpp, OpenAI, Anthropic) and Model Context Protocol (MCP/ACP) handlers natively in-process.
The Stack: Go, MCP/ACP specs, llama.cpp API, SQLite, PostgreSQL.
The Outcome: Shipped 26+ open-source releases under Apache-2.0, enabling sub-100ms local tool execution and reducing runtime footprint to a single 30MB binary.

AIAgentsGoMCPOpen Source

Contenox Teams — MCP Server & Document SaaS

TL;DR: Multi-tenant SaaS platform extending the Contenox runtime with document storage, vector search (RAG), and Model Context Protocol access.

The Problem: Teams need a centralized, secure repository to index internal documents and query them from diverse local or cloud-based AI agents without rebuilding RAG pipelines for each agent.
The Implementation: Designed a multi-tenant RAG service with document parsing and embeddings. Built the SaaS management layer, including RBAC, member invitation flows, audit logs, and Stripe billing integration.
The Stack: Go (RAG + MCP), Next.js, Postgres (pgvector), Stripe, GCP, Terraform, k3s.
The Outcome: Delivered a production-ready SaaS control plane that indexes document collections and exposes them via MCP to external agents with sub-200ms vector search latency.

SaaSMulti-tenantRAGMCPNext.js

SaaS Product

TechCareer - SaaS Architecture

TL;DR: Full-stack B2C SaaS product featuring automated CV parsing and AI-driven content generation.

The Problem: Building a high-performance consumer application that requires secure PDF handling, fast AI text generation, user onboarding, and subscription payments without infrastructure bloat.
The Implementation: Built the web application with Next.js Server Components and React. Designed database schemas and authentication policies in Supabase, and integrated Stripe billing cycles and PDF parsing utilities in serverless functions.
The Stack: Next.js, React, Supabase (Postgres), Stripe, TailwindCSS.
The Outcome: Shipped a production-ready web application with subscription payments, reducing typical development cycle to less than 3 months.

SaaSNext.jsStripeSupabase

Direktiv — Kubernetes-Native Engine

TL;DR: Event bus optimizations and observability updates for a Kubernetes-native serverless workflow engine.

The Problem: The workflow engine experienced latency spikes and event drops under high concurrent load, with insufficient runtime visibility for debugging active workflows.
The Implementation: Re-architected the internal event routing, decoupled long-running database queries, and implemented a structured slog logging system in Go. Built open-source observability pipelines using OpenTelemetry, Fluent Bit, and Prometheus.
The Stack: Go, Kubernetes, Knative, PostgreSQL, OpenTelemetry, Prometheus, Grafana.
The Outcome: Increased event processing throughput by 10x and reduced workflow execution latency on critical routing paths to under 5ms.

KubernetesGoObservabilityEvent Streaming

Multi-Tenant ERPNext Platform — Operator & Control Panel

TL;DR: Kubernetes-native provisioning platform and control plane to deploy and operate isolated ERPNext instances for B2B tenants.

The Problem: Provisioning and maintaining ERPNext instances for multiple customers required manual server configurations, leading to inconsistent configurations, slow updates, and high operational overhead.
The Implementation: Developed a custom Kubernetes operator in Go to manage the lifecycle of ERPNext containers, databases, and assets. Built a web-based administration dashboard for day-2 operations (database migrations, backups, resource scaling) and automated provisioning with Terraform and Ansible.
The Stack: Go, Python, Kubernetes (operator-sdk), Terraform, Ansible, Jenkins, IONOS Cloud.
The Outcome: Automated tenant provisioning from several hours to under 5 minutes, allowing a single operator to manage a fleet of 50+ isolated ERPNext instances.

IONOSTerraformKubernetesERPNextPlatform

Discuss a project?

Cloud architecture, IaC, or an AI-native product - I help shape production systems your team can run, with docs and runbooks.

Get in Touch