Commerce infrastructure, production delivery, and AI systems for teams that cannot afford fragile operations.
SoftSpark is a founder-led engineering company with 12+ years of operator experience. We design, build, and stabilize revenue-critical platforms. Our work stays close to production: architecture, infrastructure, releases, observability, recovery, and engineering automation.
softspark.eu · biuro@softspark.eu
SoftSpark is built and led by Łukasz Krzemień, a software and infrastructure engineer who works directly on architecture, implementation, delivery, and production operations.
Łukasz combines commerce-platform experience with infrastructure automation and AI engineering. His work spans Magento and PHP systems, cloud and VPS infrastructure, CI/CD, observability, knowledge retrieval, MCP servers, coding-agent platforms, and multi-model orchestration.
The operating principle is simple: important systems need explicit ownership, measurable acceptance criteria, repeatable delivery, and a tested recovery path.
| Area | Scope | Typical stack |
|---|---|---|
| Commerce platforms | Catalog, checkout, search, cache, PIM, integrations, multi-store operations | Magento 2, PWA, Akeneo, Symfony, Pimcore |
| Infrastructure and delivery | Provisioning, containers, CI/CD, observability, backup, recovery | Terraform, Terragrunt, Ansible, Docker, AWS, Azure, Hetzner |
| AI delivery systems | Coding-agent rules, knowledge retrieval, MCP servers, model routing, multi-agent workflows | Claude Code, Codex, Copilot, Gemini, MCP, RAG, CRAG |
| Operational enablement | SOPs, runbooks, architecture decisions, incident workflows, team tooling | GitHub Actions, GitLab CI, OpenTelemetry, Prometheus, Grafana |
We work on the application paths where latency, deployment risk, and integration failures affect revenue.
| Capability | Delivery scope |
|---|---|
| Platform engineering | Magento 2 B2C and B2B, headless storefronts, middleware, extensions, upgrade paths |
| Performance | Varnish, OpenSearch, queues, cache policy, PHP runtime, database diagnostics |
| Product data | Akeneo PIM, catalog normalization, feeds, imports, exports, synchronization |
| Operations | Multi-store releases, monitoring, incident loops, rollback procedures, peak-readiness checks |
Our infrastructure work targets repeatable changes and recoverable failures. The stability target for release-critical paths is 99.95%.
| Layer | Implementation |
|---|---|
| Provisioning | Terraform and Terragrunt for Hetzner Cloud, AWS, Azure, and hybrid VPS environments |
| Automation | Containerized Ansible, explicit inventories, idempotent playbooks, configuration ownership |
| Delivery | CI/CD validation, zero-downtime deployment patterns, health checks, rollback gates |
| Operations | Structured logs, metrics, tracing, backups, restore tests, capacity and cost reviews |
We treat AI as an engineering system, not a chat window. Rules, permissions, retrieval, evaluation, and release gates remain explicit.
| Layer | SoftSpark implementation |
|---|---|
| Engineering toolkit | ai-toolkit distributes 109 skills, 44 agents, hooks, audits, and shared policies across 12 developer-tool integrations |
| Work management | jira-mcp connects MCP clients to multiple Jira instances with ADF formatting, caching, and comment templates |
| Knowledge | Hybrid retrieval, corrective RAG, multi-hop search, reranking, evaluation sets, and knowledge-gap tracking |
| Model access | dsh-codex connects DeepSeek Harness to a locally authenticated Codex app server using a ChatGPT subscription |
| Orchestration | dsh-orchestrator delegates bounded tasks to Claude Code and GitHub Copilot Gemini through native subscription logins |
| Control plane | Approval gates, least-privilege tools, credential boundaries, SARIF audits, signed package provenance, human review |
- Diagnose. Inspect architecture, release flow, runtime evidence, operational risk, and the actual bottleneck.
- Design. Define ownership, interfaces, environments, migration steps, failure modes, and measurable acceptance criteria.
- Codify. Implement infrastructure, services, pipelines, policies, tests, SOPs, and runbooks.
- Ship. Release small reversible changes with automated gates, health verification, and a rollback path.
- Transfer. Leave the team with documented systems, maintainable tooling, and evidence they can reproduce.
- Stabilize Magento, PWA, search, cache, queue, or integration paths that affect conversion and release safety.
- Standardize infrastructure that has grown through manual changes and undocumented dependencies.
- Build CI/CD, observability, backup, restore, and incident procedures around an existing production platform.
- Convert AI experiments into governed tools with retrieval, permissions, evaluation, audit trails, and human approval.
- Create reusable engineering systems: modules, MCP servers, agent workflows, SOPs, runbooks, and knowledge bases.
| Project | Purpose | Distribution | License |
|---|---|---|---|
| ai-toolkit | Shared skills, agents, hooks, policies, audits, and cross-tool generators |  |
Apache-2.0 |
| jira-mcp | Multi-instance Jira MCP server with ADF conversion and local task caching |  |
Apache-2.0 |
| dsh-codex | DeepSeek Harness provider for locally authenticated Codex app-server sessions |  |
Apache-2.0 |
| dsh-orchestrator | DeepSeek Harness bundle for Claude Code and Copilot Gemini delegation |  |
Apache-2.0 |
Every public module starts at 1.0.0, ships with Apache-2.0 and NOTICE, and uses automated supply-chain verification appropriate to its distribution channel.
Architecture review, commerce stabilization, infrastructure automation, or AI delivery system:
softspark.eu · biuro@softspark.eu
This profile is licensed under Apache-2.0. See NOTICE. Linked repositories carry their own licenses.