Our Work

Selected Engineering Work

Oronts products, open-source tools and the production AI system running on this site. Everything here is our own engineering: inspectable, documented, and built with production discipline from day one.

7Engineering projects
2Open source on GitHub
3Project categories
2024-25Active years
How We Work

Engineering discipline before everything else

  • 01Every engagement starts with a written architecture decision record. We document trade-offs, alternatives considered, and reasons for each technical choice before writing code.
  • 02No feature ships without a defined rollback path, observability instrumentation, and load testing under realistic conditions.
  • 03Production readiness reviews gate every release. We audit security boundaries, failure modes, data integrity guarantees, and operational runbooks before go-live.
  • 04We open-source reusable components wherever client agreements allow. Our open-source tools run in production alongside the systems they were extracted from.
Read the full case
Production PlatformActive

Exfinity

A multi-tenant AI commerce platform for the experiences and ticketing industry. Merchants connect once and sell experiences from many suppliers in natural language, through unified search, AI booking agents and embeddable widgets, all under their own brand. Built for Exfinity GmbH.

2025Ongoing since 2025
Exfinity GmbH
8Specialized AI agents
Live DemoRead Guide
Discuss a Similar Project
The Challenge

The experiences market is fragmented. Every merchant rebuilds the same supplier connections from scratch, supplier catalogs arrive in mismatched formats, and no one can offer AI-powered search or booking without first building their own AI systems. There is no single, safe way to search across every supplier or let an AI agent book on a customer's behalf.

Our Approach

Exfinity works like a Stripe for experiences. A merchant connects once and gets search, booking and support across every supplier, all under their own brand. Behind that simplicity, supplier catalogs are pulled in automatically, cleaned into one common format and made searchable by meaning, not just by keywords. A team of eight AI agents handles discovery, booking and support, and they can act only through a controlled set of tools with strict per-merchant rules, rate limits and a full audit trail. Everything runs in the EU, in Frankfurt, and personal data is masked before it ever reaches an AI model.

The Result

One integration gives a merchant AI-powered search, booking and support across every connected supplier. It is built to production standards from day one: fast search, strict separation between each merchant's data, and a learning loop that turns weak answers into reviewed knowledge over time. Designed and built for Exfinity GmbH since 2025.

Read the full case
Open SourceActive

Vendure Data Hub

Enterprise ETL and data integration plugin for Vendure. Visual pipeline builder, 9 extractors, 61 transform operators, 24 entity loaders, and feed generators for Google Merchant and Amazon.

2024Ongoing
Open source
9Extractors (verifiable in the repository)
GitHubRead Guide
The Challenge

Vendure projects keep rebuilding the same plumbing: product imports from ERP and PIM systems, inventory sync, price updates, marketplace feeds. Each integration starts from zero, ships as one-off scripts, and breaks silently when a supplier changes a column. The ecosystem had no production-grade, reusable data pipeline layer.

Our Approach

We built Data Hub as a first-class Vendure plugin: declarative pipelines composed from extractors (CSV, JSON, XML, REST, GraphQL, FTP, S3 and more), 61 transform operators with dry-run preview, and loaders for 24 Vendure entity types. Pipelines run on schedules or webhooks, with retries, idempotent upserts, real-time logs and a visual editor in the admin UI. Feed generators publish Google Merchant and Amazon feeds from the same pipeline graph.

The Result

A single plugin replaces the integration scripts of a typical commerce project. Published open source on GitHub; production-tested with high-volume catalog imports and verifiable down to every operator in the repository.

Read the full case
Oronts ProductActive

OGuardAI

Semantic data protection runtime for AI systems. Policy engine, PII detection and reversible tokenization between your application and any LLM.

2024Ongoing
Oronts GmbH
99.7%PII detection recall on our test corpus
Live DemoRead Guide
Discuss a Similar Project
The Challenge

Production LLM systems share a failure class: PII leaking into prompts and outputs, hallucinated data reaching customers, and generated text violating communication policies. Ad-hoc filters solve one incident and break on the next edge case. GDPR makes this an architecture problem, not a patching problem.

Our Approach

OGuardAI runs as a synchronous filter in the LLM request and response path. Three validation layers: content policy classification against hot-reloadable YAML rule sets, PII detection combining pattern matching with named entity recognition, and semantic tokenization that swaps sensitive values for reversible tokens so the model never sees raw data. Restoration happens per output channel policy.

The Result

A reusable, framework-agnostic guardrail layer that turns GDPR-aware AI from per-project firefighting into infrastructure. A proprietary Oronts product, self-hosted and provider-neutral, with the architecture documented in our data-leakage guide.

Read the full case
Production PlatformActive

Oronts AI Assistant

The production Mastra agent system answering on this site: five agents, four tools, streaming responses, and a hardened API layer. Try it in the corner of this page.

2025Ongoing
Oronts GmbH
5Specialized agents in production
Live DemoRead Guide
Discuss a Similar Project
The Challenge

AI capability claims are cheap. The honest proof is an agent system running in production, with the same constraints clients face: rate limiting, CSRF, prompt injection surface, cost control, observability and a real lead pipeline behind it.

Our Approach

Built on Mastra with five specialized agents (quote, chat, vision, voice, analytics) and four tools that structure conversations into qualified leads: proposal submission, call scheduling, email drafting and conversation summaries. Responses stream over the Vercel AI SDK. The API layer applies rate limiting, CSRF, origin checks and request tracing; tool calls route through an authenticated internal proxy into the Oronts API. The agent never invents prices: it qualifies, structures and hands over to a human.

The Result

A working answer to the question every buyer should ask an AI vendor: show me yours. The assistant on this page is the deployment; its design decisions are documented in our guides.

Open SourceActive

Pimcore Asset Pilot Bundle

Rule-based digital asset organization for Pimcore: configurable rules, priority ordering and audit logging for libraries that grow faster than teams can sort them.

2025Ongoing
Open source
100%Actions captured in the audit log
GitHub
The Challenge

Large Pimcore installations accumulate tens of thousands of assets. Editors drop files wherever upload dialogs open; naming conventions drift; the DAM becomes a junk drawer. Manual cleanup does not scale and one-off migration scripts rot.

Our Approach

Asset Pilot applies declarative organization rules to the asset tree: match conditions on metadata, file type, naming patterns or upload context, then actions like move, rename, tag or assign. Rules carry priorities and run on upload events or as batch jobs over existing libraries. Every action lands in an audit log, and a dry-run mode previews the impact before anything moves.

The Result

Asset chaos becomes a configuration problem instead of a recurring cleanup project. Published open source on GitHub.

Oronts ProductActive

Vendure Customer Intelligence

Customer engagement plugin for Vendure: wishlists, reviews, loyalty points and cart recovery as one coherent domain instead of four bolted-on apps.

2025Ongoing
Oronts GmbH
4Engagement domains in one plugin
Read Guide
Discuss a Similar Project
The Challenge

Shops assemble engagement features from disconnected plugins: one for reviews, one for wishlists, a third for loyalty. Each has its own data model and admin UI; none shares a customer picture. Marketing then exports CSVs to guess at segments.

Our Approach

One plugin, one customer-engagement domain. Wishlists, verified-purchase reviews, a configurable loyalty engine and abandoned-cart recovery share entities, events and admin screens. Everything emits Vendure events, so segments and automations build on real-time signals instead of exports.

The Result

A single engagement layer with a unified customer view, built on Vendure's plugin architecture and event bus. Oronts product, documented in the portfolio.

Have a project in mind?

We take on selected engagements each quarter. Tell us what you are building.