Advice where you need it, engineering all the way to production. Not slide decks: working systems on your data, with a definition of done and a test to prove it.
Assistants over your own documents that answer with the exact passage they used, so every answer can be checked, and say so when the files do not cover a question.
Agents that plan, call your tools and finish multi-step tasks, with guardrails and a human in the loop wherever the decision matters.
Model Context Protocol servers that give AI models governed access to your databases and internal tools through one standard, without one-off integrations.
Pipelines that classify, extract and route what used to be handled by hand: reports, documents, tickets, emails.
Models and dashboards that turn operational data into forecasts and decisions your team actually uses.
Benchmarks with held-out test sets, LLM grading you can audit, and fine-tuning when the data justifies it, so you know how good a system is before it ships.
Two products of our own, built to the same standard as our client work: one for companies that need their knowledge answered with proof, one for anyone with a Mac and a TV.
Your knowledge, mastered.
Meistra reads every file your company has, writes down what it learned with the page it came from, and trains an assistant that must pass a test on your files before it answers anyone.
Your Mac screen, on any screen in the house.
Broadcast your Mac screen, with system audio, to any browser on your Wi-Fi: laptops, phones, or a Fire TV. Free, open source, nothing to install on the viewing device.
AI agents, private RAG, MCP servers and automation, built for clients and running in production.
An AI agent that answers customers on WhatsApp and email in their own language, follows the shop’s support playbooks, looks up live orders through a Shopify MCP integration, and runs in autopilot or assisted mode per conversation.
A Model Context Protocol server on top of Dremio that lets AI agents answer natural-language questions over enterprise datasets, with the same permissions and audit trail as a human analyst.
A policy-aware assistant that answers employee questions from the current HR policies and cites the exact paragraph, with role-based access and versioned documents.
'/%3e%3ccircle%20cx='980'%20cy='120'%20r='260'%20fill='%2310A89A'%20opacity='0.08'/%3e%3cg%20transform='translate(150%20120)%20rotate(-6)'%20filter='url(%23sh)'%3e%3crect%20width='300'%20height='400'%20rx='18'%20fill='%23FFFFFF'/%3e%3crect%20x='34'%20y='40'%20width='130'%20height='14'%20rx='7'%20fill='%23B3C7D9'/%3e%3crect%20x='34'%20y='78'%20width='232'%20height='10'%20rx='5'%20fill='%23E1EAF2'/%3e%3crect%20x='34'%20y='102'%20width='200'%20height='10'%20rx='5'%20fill='%23E1EAF2'/%3e%3crect%20x='34'%20y='126'%20width='220'%20height='10'%20rx='5'%20fill='%23E1EAF2'/%3e%3cg%20transform='translate(34%20190)'%3e%3crect%20x='0'%20y='80'%20width='30'%20height='60'%20rx='6'%20fill='%23B3C7D9'/%3e%3crect%20x='44'%20y='40'%20width='30'%20height='100'%20rx='6'%20fill='%237A9BB8'/%3e%3crect%20x='88'%20y='60'%20width='30'%20height='80'%20rx='6'%20fill='%23B3C7D9'/%3e%3crect%20x='132'%20y='10'%20width='30'%20height='130'%20rx='6'%20fill='url(%23teal)'/%3e%3crect%20x='176'%20y='30'%20width='30'%20height='110'%20rx='6'%20fill='%237A9BB8'/%3e%3c/g%3e%3c/g%3e%3cg%20transform='translate(560%20260)'%3e%3cpath%20d='M0%2024h150'%20stroke='%2310A89A'%20stroke-width='6'%20stroke-linecap='round'%20stroke-dasharray='2%2022'/%3e%3cpath%20d='M150%2024l-22-16m22%2016l-22%2016'%20stroke='%2310A89A'%20stroke-width='6'%20stroke-linecap='round'%20stroke-linejoin='round'%20fill='none'/%3e%3c/g%3e%3cg%20transform='translate(760%20110)%20rotate(4)'%20filter='url(%23sh)'%3e%3crect%20width='320'%20height='420'%20rx='18'%20fill='%23111D35'/%3e%3crect%20x='36'%20y='44'%20width='90'%20height='12'%20rx='6'%20fill='%232EC4B6'/%3e%3crect%20x='36'%20y='80'%20width='200'%20height='18'%20rx='9'%20fill='%23FFFFFF'%20opacity='0.9'/%3e%3crect%20x='36'%20y='124'%20width='248'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='148'%20width='230'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='172'%20width='248'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='196'%20width='180'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='240'%20width='120'%20height='12'%20rx='6'%20fill='%232EC4B6'%20opacity='0.7'/%3e%3crect%20x='36'%20y='272'%20width='248'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='296'%20width='210'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='320'%20width='236'%20height='10'%20rx='5'%20fill='%234A7099'/%3e%3crect%20x='36'%20y='364'%20width='110'%20height='30'%20rx='15'%20fill='%2310A89A'/%3e%3c/g%3e%3c/svg%3e)
An agent pipeline that turns structured financial data and KPIs into narrative reports: executive summary, analysis sections and recommendations, in the client’s own template.
Dataotter is an AI consultancy and engineering company in Nuremberg, Germany, founded in 2025. We advise companies in Germany, Austria, Switzerland and the rest of the EU on where AI pays off, build the systems, and run them in production. We also ship products of our own.
We work with companies that have real data and real constraints: policies that must be cited, numbers that must be right, systems that must survive contact with users. Our way of working is small teams, short iterations and a clear definition of done.
Our products come out of that work. Meistra is what we think every AI assistant should do: read everything, cite its sources, and pass a test before it goes live. Mac Screencast is the tool we wanted for our own living rooms, released as open source.
Advice on where AI pays off, then the engineering to get it live: for clients, and for our own products Meistra and Mac Screencast.
Answers with sources, held-out test sets, and a report on how good a system is before it ships.
EU hosting, data processing agreements, and architectures that keep your files where they belong.
Most AI projects prove that the model works, then stop. We build for the part after the demo.
Today's models are capable enough. What fails is integration, permissions, evaluation and ownership. We plan those from day one.
Short iterations with a definition of done: what the system must answer, how it is graded, who signs off. A pilot ends with a decision, not a slide deck.
Every answer links to its source, every action leaves a trace. In legal, HR and compliance that is the difference between a tool people use and one they switch off.
The systems we build draft, retrieve and recommend. People keep the decision, and the assistant earns trust topic by topic.
On Agentic Fine-Tuning, RAG in production, and moving AI from pilot to production.
Companies sit on decades of data whose lessons no human has ever read in full. A walk through Agentic Finetuning, the training loop that extracts those learnings, proves them, and keeps them alive.
Read on MediumA practical walkthrough of fine-tuning BERT with the Ktrain library to classify disaster-related tweets, from data preparation to evaluation.
Read on MediumUsing NLP and machine learning to predict a patient’s medical condition from the reviews they write about their medication. Part one of a healthcare series.
Read on MediumTell us what you want to automate or answer. We reply within one working day.
We look at your data, your process and what a good answer would be.
Scope, definition of done, timeline and price. Usually within a week.
A working system on your data, tested against agreed questions. Weeks, not quarters.