References

Several systems live. We are allowed to name four of them.

Every project on this page runs with real users under real constraints: a legal department, a hosting provider’s support queue, an insurance data pipeline, a company switchboard, two sales organisations with nationwide installer networks. Some of it we can only describe anonymised, because that was the condition for building it at all.

Delivered
70+ AI projects across eleven industries
Named here
Six, two anonymised at the client’s request
In use
With real users under real-world conditions
Type
6 references
Bespoke AI Project
Public broadcasting · Legal · Anonymised

National Public Broadcaster

Production LLM pipeline for legal document automation

Case summaries from a containerised LLM pipeline, each with full source citation.

Text ExtractionQualitative AssessmentContent Generation
Bespoke AI Project
Web Hosting · IT Services

webgo GmbH

On-premise RAG pipeline for AI-supported customer service

Built on webgo's own infrastructure under strict data sovereignty.

Knowledge RetrievalContent Generation
Bespoke AI Project
Insurance · M&A

Tjara GmbH

Cloud SaaS platform for automated data-quality scoring

Multi-stage LLM scoring and a GDPR-compliant data flow.

Structured EnrichmentPredictive Scoring
Bespoke AI Project
Enterprise Communications · SaaS

CyberTwice B.V.

AI voicebot framework as digital reception

A modular voicebot framework deployed on Azure.

Conversational AgentSpeech Synthesis
Product Solution
Heating technology · Trade channel sales · Anonymised

Heating technology manufacturer

Field Sales Radar for market coverage across the trade channel

Around 80,000 installers identified nationwide, qualified by offering and competence, and reconciled against the existing CRM. Delivered as an import-ready export back into the client's own system, repeated annually since 2023.

Market IdentificationPredictive Scoring & ClassificationCRM Enrichment
Product Solution
DIY retail · Solution business · Anonymised

Major DIY retail chain

Field Sales Radar for partner coverage in the solution business

Landscaping and outdoor-build trades identified, qualified by degree of professionalisation, and matched against the existing partner base. Up to six parallel sources were merged into one record per business, traceable back to where each entry came from.

Market SegmentationRecord ConsolidationPredictive Scoring & Classification

Six industries, six stacks, very different budgets. The same four conditions decided whether each one made it into production.

01

The data never left where it was allowed to be.

Broadcaster: anonymised. webgo: entirely on their own hardware. Tjara: a GDPR-compliant flow designed before the first prototype existed. In each case the deployment model was a requirement rather than an afterthought, which is also what narrowed the model choice.

02

A human stayed in the loop where it mattered.

None of these systems decides on its own. They summarise, score, retrieve and draft; a specialist accepts, corrects or rejects. That review step is also where the labelled cases came from that we used to widen coverage.

03

Every output can be traced back to a source.

Case summaries cite the passage they came from. Retrieval answers name the document. Scores expose the features behind them. Without that, nobody in a regulated department signs off on day-to-day use.

04

Someone owned it after go-live.

Each project ended with a named owner on the client side, a written operating model and monitoring they can read themselves. The projects that quietly stall are the ones where this last step was treated as optional.

The stack behind our projects.

The tools and platforms behind every Blue Avenir engagement.

Deployment model

Cloud, private cloud, or entirely on your own infrastructure. The webgo pipeline runs on-premise under strict data sovereignty.

  • On-premise
  • Private cloud
  • AWS
  • Azure
Data protection

GDPR-compliant data flows are designed in from the first prototype, not retrofitted before go-live.

  • DE/EU data residency
  • Self-hosted models
  • Audit trails
  • Role-based access
Data & storage

Where state, embeddings and artefacts live.

  • PostgreSQL
  • Redis
  • S3
  • MongoDB
Models

Chosen per use case, and replaceable. No architecture is tied to a single provider.

  • OpenAI
  • Anthropic
  • Google Gemini
  • Mistral
  • Ollama
  • Microsoft Phi
Retrieval & orchestration

How a model reaches your knowledge, and how the surrounding steps are sequenced.

  • RAG
  • Vector databases
  • MCP
  • LangChain
  • Airflow
  • n8n
Runtime & delivery

What the running system is built and shipped with.

  • Python
  • FastAPI
  • Docker
  • Kubernetes
  • Next.js
  • TypeScript

From pilot to production, together.

We help you close the gap between a promising pilot and a system your teams rely on every day.