Our verdict
dida builds tailor-made machine learning software, says it has deployed most of it in production, and in October 2026 its consortium won first place in the first evaluation round of the German Cyberagentur’s HEGEMON project on foundation models for security use (homepage; dida). It suits a German company or public agency with a hard language, vision, or document problem that off-the-shelf tools cannot solve, and it is less suited to a buyer that wants a large team or a quick packaged rollout.
The right buyer
- Industrial companies with complex customer service. For nVent, an electrical products maker with 11,000 employees, it built a multi-agent system that drafts fact-checked replies to customer emails and Salesforce tickets with source citations (nVent project).
- German public agencies. It built a semantic search that matches free-text business descriptions to more than 800 industry codes for a North Rhine-Westphalia service portal run by PublicPlan (public administration project).
- Buyers who need research-grade methods. Its team publishes papers, and its co-founder presented two diffusion-sampling papers at ICML 2026 (dida; about page).
Reasons to look elsewhere
- With about 40 people, it is a small team for a broad enterprise program.
- Most of its cases and clients are in Germany.
dida, At A High Level
dida Datenschmiede GmbH was registered in Berlin in March 2018 to provide IT services and software development, especially in machine learning (North Data). It is led by managing directors Lorenz Richter, a mathematician and its co-founder and chief technology officer, and Markus Düttmann, a physicist who previously founded the ad-buying platform realzeit; co-founder Philipp Jackmuth, a statistician, advises clients on which processes to automate (about page; North Data).
- Offices at Ritterstraße 12-14 in Berlin, with some staff working remotely from other European countries (about page).
- 41 employees in its 2024 accounts; we estimate annual revenue at $5 million to $7 million (Gründungsindex).
- Has received more than a dozen public research grants, including one of €489,523 in 2025 for explainable AI in legal retrieval systems (North Data).
- Services cover machine learning solutions, consulting, MLOps, and research, often in natural language processing and computer vision (homepage).
Scores by criterion
We scored dida from its homepage, about, project, and news pages and from its commercial register filings. Technical depth scores highest; commercial clarity scores lowest because it publishes no terms.
| Criterion | Weight | Score | Basis |
|---|---|---|---|
| Production track record | 20% | 6.5 | Published production projects for nVent, PublicPlan, and felmo; says most of its software runs in production. |
| Technical depth | 15% | 8.0 | Research team with many PhDs; ICML papers; LLM agents, NLP, and vision. |
| Industry depth | 10% | 6.0 | Industry, public sector, and remote sensing cases. |
| Delivery model and handoff | 15% | 6.5 | End-to-end from problem definition and planning to deployment and MLOps. |
| Commercial clarity | 10% | 5.0 | No published rates or minimums. |
| Platform independence | 10% | 7.0 | Custom models; no tie to a single platform. |
| Stability and scale | 10% | 5.5 | Founded 2018; 41 employees. |
| Independent evidence | 10% | 7.0 | HEGEMON first place, public research grants, and client reviews. |
| Overall | 100% | 6.5 | Weighted average |
Where dida fits
dida is used by German companies, public agencies, and research consortiums that need custom machine learning for documents, language, images, or geodata, built by a team that can read and write the research. On price it sits with specialist German AI boutiques, above offshore teams and below large consultancies. Competitors on this site include ML6 and Unit8 for European AI engineering, and deepsense.ai for research-heavy machine learning work.
| Need | Fit | Why |
|---|---|---|
| LLM agents for technical customer service | Strong | It built nVent’s multi-agent reply system with source citations. |
| Language or document AI for a German agency | Strong | It has public sector projects and won first place in HEGEMON’s first round. |
| Computer vision for manufacturing or remote sensing | Good | Its projects include defect detection and mine detection from imagery. |
| A large multi-team enterprise rollout | Weak | Its team of about 40 is small for that scale. |
What clients and leaders say
dida’s leaders and clients describe its work in published announcements. One buyer concern is written from public facts about the firm.
“We were very impressed with their deep technical knowledge and their ability to explain it in plain terms.”
“Their NLP knowledge and expertise are impressive and it really made a difference.”
“Their domain expertise in different machine learning processing models was elemental to the success of our project.”
“Their engineers solved a problem our tools could not, but with a team of about 40 we had to plan around their availability, and nothing about pricing was published.”
Where customer reviews couldn’t be found, our analysts attempted to give a voice to buyer sentiment, drawing on real information from review sites, forums and product documentation. Based on: dida’s 2024 company data, which lists 41 employees, and its homepage.
Who leads dida
Lorenz Richter co-founded dida, leads its machine learning team, and is a managing director (about page; dida).
Sources (8)
- dida: about us
- dida: homepage
- North Data: dida Datenschmiede GmbH register filings
- Gründungsindex: dida Datenschmiede GmbH company data
- dida: wins first rank in Cyberagentur’s HEGEMON evaluation (October 2026)
- dida: AI agents for industrial customer service (nVent)
- dida: semantic search for public administration
- dida: Lorenz Richter presents two papers at ICML (August 2026)