Skip to content

APPLIED AI · DATA ENGINEERING

The signal is already in your data.

3 project spots open for 2026

We build the data foundation and the AI that runs on it: retrieval, agents, and private models on your own infrastructure. Designed, built, and maintained by the same people.

Book a callSee what we build

Most companies sit on years of data their AI can't touch. It's scattered across file shares, databases, wikis, and inboxes, and no off-the-shelf tool has ever seen it. We build the layer in between. Systems that retrieve what your business already knows, reason over it, and act on it, without handing anything to someone else's cloud. And when the foundation isn't there yet, we build that first.

SERVICES · 4 LAYERS

What we
build

Four layers, in the order most projects run. Build the data foundation and turn it into answers. Make it findable. Put AI to work on it. Keep all of it inside your walls. We start from whatever you have, an existing platform or a blank slate.

Field notes
01

Data Foundation & Analytics

PIPELINES · ANALYTICS · GOVERNANCE

The groundwork everything else runs on. Data that is clean, modeled, and governed, and the analytics that turn it into decisions.

  • Data pipelines and ELT, batch and streaming, orchestrated and version-controlled
  • Warehouses, lakehouses, and the storage and compute architecture underneath
  • Data modeling, schemas, and data contracts that keep sources consistent
  • Catalog and lineage, so teams can find what exists and trust where it came from
  • Quality checks on every run, with monitoring, alerting, and clear SLAs
  • Access control, PII handling, encryption
  • Migrations off legacy systems, run beside the old platform until the numbers match
  • Metrics layers and semantic models so the whole company measures the same thing
  • BI and dashboards built on definitions people trust
  • Forecasting and statistical modeling where it changes a decision
02

Retrieval & Knowledge

EMBEDDINGS · HYBRID SEARCH · GRAPH RAG

We turn what you have into a corpus a model can answer from.

  • Ingestion that turns file shares, exports, wikis, and archives into one clean, queryable corpus
  • Semantic search: an encoder embeds your documents, a vector database serves the nearest neighbors in milliseconds
  • Hybrid retrieval, BM25 and vectors scored together, because people search in exact part numbers as often as in sentences
  • Graph RAG where the answer lives in relationships: entities extracted from your data, linked, and traversed at query time
  • Rerankers that cut the candidate set down to the few passages worth a model's attention
  • Embedding pipelines that keep the index current as documents change, without full re-runs
  • Structured extraction from messy and mixed-format sources
03

Agents & Automation

AGENTS · MCP · WORKFLOWS

Systems that act on your data, not just answer questions about it.

  • Autonomous and multi-agent systems with real tool use, step limits, and guardrails
  • Custom MCP servers that expose your databases, APIs, and file stores to any model that speaks the protocol, under permissions you set
  • Workflow automation that wires models into the stack you already run
  • Pipelines your non-technical teams drive in plain language, no scripts, no ticket queue
  • Evaluation and observability, so what worked in the demo still works in month six
04

Private & Owned

ON-PREM · OPEN WEIGHTS · FINE-TUNING

Models that run inside your network. Your data never leaves it.

  • Open-weight LLMs deployed on your hardware, from a single GPU box to a cluster
  • Fine-tuning and training on your own data, for your own domain
  • The right architecture for the job: encoders for embeddings and classification, decoders for generation
  • Private-by-design architecture for teams that can't send data to a public API
  • Prompt and system design that gets real work out of the models you own

PROCESS

How it
works

  1. 00

    Meet

    We come to you for two hours, on-site, and go through your data and systems together. Requirements, timelines, expectations. Then you get it in writing, including what we would not build. Free, and without obligation.

  2. 01

    Map

    Together we turn that into architecture, data contracts, and the exact systems to build. Scoped and sequenced so the work that pays off first comes first.

  3. 02

    Build

    Then we build. Pipelines, retrieval, agents, models, the analytics on top, whatever the plan calls for, and the unglamorous parts too: auth, monitoring, docs.

  4. 03

    Ship & run

    It goes live in your environment, documented and owned by you. Then we stay. Keeping systems working is the part of this business most firms quietly skip. We scope it into every project and stay as your data changes shape.

WHO YOU'RE WORKING WITH

  • VIENNA · AT
  • SMALL ON PURPOSE
  • END-TO-END
  • A FEW CLIENTS AT A TIME

We're a small AI and data engineering studio in Vienna, small on purpose. The people who design your architecture also write the code and answer when something breaks. Your project never gets handed down a chain.

We build end to end: the data platform, the retrieval and agents on top of it, the deployment on your infrastructure. This is hands-on engineering. You end up with production systems you own, not a strategy deck.

If your data is complex and your requirements are real, that's the work we want.

FAQ

Questions we get

Where does our data live? Does it leave our network?

It doesn't leave. Retrieval, agents, and models run on your infrastructure, on-prem or in your private cloud, and the default is that nothing goes to a public API. Where a design would genuinely benefit from an external model, we say so up front and you decide.

Do you only work with data we already have, or can you build from scratch?

Both. If the foundation is there, we build on it. If it isn't, we build the foundation first.

Is data engineering and analytics part of this, or just AI?

Both. Data engineering and analytics are half of what we do: pipelines, warehouses, governance, metrics, BI. The AI layer is built on that work, and that work is usually why the AI works.

Is the first visit really free?

Yes, genuinely. The two hours and the written plan that follows cost nothing. We do it because it's the fastest way for both sides to find out whether there's a real project here.

What happens after it ships?

We stay on. Pipelines drift, sources change, models age, and someone has to be watching. Maintenance is scoped into every project from the start.

Who does the work?

The people you meet in the first two hours. We're deliberately small and take on a few clients at a time; nothing gets delegated to a bench we don't have.

What does a project cost?

The honest answer is that it depends, and the free visit exists so neither of us has to guess. The plan it produces lays out scope, sequence, and effort, and you see the number before you commit to anything.

Which models do you use?

Whichever fits the task and the constraint. Open-weight models deployed inside your network when data can't leave, hosted models where it can. Encoders for search and classification, decoders for generation. We resell nothing and take no referral fees, so the recommendation is only ever the recommendation.

We already use Claude or ChatGPT. Where do you fit?

Those assistants only know what they can reach. We build custom MCP servers that connect them, or any model that speaks the protocol, to your databases, documents, and internal APIs, with permissions you control and an audit log you keep. Same assistants, suddenly useful on your data.

THE FIRST STEP · FREE · ON-SITE

The first two hours are on us.

We come to you. Two hours on-site, in person, walking through your data, your systems, and what's worth building. Afterward we write it up: an implementation plan covering what to build, in what order, and what it takes to run. The plan is yours to keep whether or not we ever work together.

Book the walk-through
  • 2 HOURSOn-site, in front of your real systems
  • IN WRITINGAn implementation plan you could hand to any engineering team
  • €0Free. No obligation, and no sales follow-up unless you ask.

Have data your AI can't use yet?

Tell us what you're sitting on. We'll give you a straight technical read on what's possible. No pitch.

Book a callor email hello@rushb.co