Knowledge assistant / internal search
Answers with source citations from your scattered knowledge base - manuals, wikis, Confluence. Hybrid search and re-ranking deliver real hits instead of word salad.
Proof of Concept (PoC) for RAG & AI systems - a solid decision in 1-3 weeks
Instead of months of upfront investment, we build something that actually runs on your data in a tightly scoped sprint and measure its quality. The result is not a glossy demo but an honest answer: viable or not - backed by measured numbers and a clear recommendation.
A paid, tightly scoped sprint is not a cost item but the cheapest insurance against an expensive misinvestment.
A test of EUR 5,000 to 10,000 before you put EUR 50,000 or more into a production system. You are buying certainty, not hope.
You receive measured numbers and a clear recommendation - a solid basis for internal sign-off, instead of gut feeling versus gut feeling.
A fixed-price sprint commits real data and real time - yours as well as ours. That produces results on real data instead of polished demo effects from a free pitch.
Four typical starting points where a retrieval-based AI system can be validated quickly and measurably.
Answers with source citations from your scattered knowledge base - manuals, wikis, Confluence. Hybrid search and re-ranking deliver real hits instead of word salad.
Make contracts, policies and long PDFs structurally searchable - with cited answers instead of made-up summaries.
An assistant that suggests answers to your team and links sources - connected to your systems via MCP, with a human as the final authority.
Evaluate PDFs, tables and scans together. Agentic retrieval specifically improves hit quality here - as a retrieval step, not as an autonomous agent.
The more data and questions, the more reliable the result. Each tier has a fixed timebox and a capped data volume - no hidden items.
Fixed timebox, capped data volume.
Fixed timebox, capped data volume.
Fixed timebox, capped data volume.
All prices are fixed prices - and can be partially credited against a subsequent follow-up project. The sprint is therefore not a sunk cost but a paid, low-risk entry point.
Honest from the start: a sprint is a proof of feasibility, not a production system.
Together we sharpen the use case, the success criterion and the timebox - so the sprint targets what matters.
You provide the relevant data in machine-readable form. On request, we clarify the data processing agreement beforehand.
Ingestion, embedding, vector database and retrieval - the working core of the system is built on your data.
Measurement against a jointly defined question set: precision/recall, hallucinations, cost - numbers instead of demo feeling.
You receive the measured results, a clear recommendation and - depending on the tier - a roadmap for the next steps.
We know the path from demo to operation - including the points where simple RAG setups usually fail.
A multimodal search index across 360,000+ objects - live in production, not a prototype gathering dust.
A custom embedding model trained on the domain - +27 percentage points of retrieval quality over standard embeddings.
Measured precision/recall instead of demo effects - quality is quantified, not claimed.
With cost guardrails and provider abstraction - spending stays predictable and the provider stays replaceable.
To make the sprint reliable, we need your input in a few places.
Even a well-founded 'not viable this way' is a valid, valuable result - you avoid a misinvestment. In any case you receive measured numbers and a clear recommendation.
We clarify the data processing agreement in advance, and the PoC data is deleted after completion. On request, the sprint runs GDPR-compliant with EU hosting or on-premise.
Yes. If you commission a follow-up project afterwards, we credit the PoC amount on a pro-rata basis.
A feasibility sprint may qualify as an eligible consulting service (e.g. under German BAFA funding). Eligibility is assessed individually - we will tell you, without obligation, what is realistic.
The audit evaluates an existing system. The feasibility sprint builds something that runs on your data and measures its quality - for the case where you don't have a system yet.
Depending on the tier, 1-3 weeks from data provisioning. You provide the data in machine-readable form, a subject-matter contact and a question set with real questions.
A feasibility sprint costs a fraction of what a failed AI project consumes. Let us talk about your use case.