Intelligent knowledge,
directly in your practice.
Guidelines, treatment schemes and price lists are uploaded, automatically understood and searchable in seconds — with four parallel retrieval strategies. Fully local, no cloud, with traceable source references.
Upload. Ask.
Reliable answer.
Three steps. No manual tagging. No cloud transfer.
Upload document
PDF, guideline, scheme or price list. The system recognises content automatically — terminology, products, diagnoses and their relationships are extracted.
Ask a question
Four strategies run in parallel: semantic search, keyword search, graph search and temporal weighting. Results are merged and re-ranked.
Answer with sources
Every answer is linked to the documents used. Traceable, transparent and verifiable — a prerequisite for medical usability.
Eight capabilities.
One combined search.
Classical AI knowledge systems use a single search path. Aescuris combines eight specialised methods into one result.
Semantic search
Understands the meaning of a question — not just individual words. "Wound not healing" also finds "stagnation in wound healing", without those exact words being used.
Keyword search
Reliably finds exact terms such as PZN numbers, product names or ICD codes — even where semantic search would miss them because no meaning trace can be detected.
Graph search
Recognises connections between concepts: "fibrin coating" linked to "phase 3" linked to "foam dressing". Cross-references between documents become visible automatically.
Temporal search
Considers temporal relevance. Current guidelines are weighted higher than older versions. Seasonal patterns are recognised.
Result fusion
Hits from all four search paths are merged. A result that appears in several strategies simultaneously receives a higher weighting.
Cross-encoder reranking
The best results are then reviewed by a specialised scoring model that analyses question and answer together. Relevance improves noticeably.
Insight derivation
New insights are derived from stored knowledge — even for questions that are not explicitly answered in any single document, such as patterns in treatment courses.
Automatic extraction
On upload, the system identifies terminology, products, active substances, diagnoses and their relationships. Manual tagging is eliminated entirely.
Classical retrieval
versus Aescuris.
The difference between a single search trace and a layered approach becomes evident quickly in practice.
One search path
- Exact terms such as PZN or product names are poorly captured
- Connections between documents remain invisible
- No re-evaluation of relevance after the first search
- No consideration of source recency
Four parallel paths
- Captures meaning, exact terms and cross-references at the same time
- Recognises relationships between documents and concepts
- Re-evaluates each result in question-and-answer context
- Considers guideline recency and seasonal patterns
One space per module.
Clear borders, clear answers.
Every Aescuris module operates on its own isolated knowledge space. A question about wound care will never inadvertently reach diagnostic documents.
What happens
automatically on upload.
Each document passes through a multi-step process — automatically, without manual preparation.
Upload
A PDF is selected via the dashboard — a guideline, a treatment scheme or a price list.
Text extraction
AI-powered text recognition reads even complex documents fully — including tables.
Segmentation
Content is broken into meaningful sections. Chapter structure and prose remain intact.
Enrichment per section
- aVectorisation as semantic representation
- bExtraction of terminology, products and diagnoses
- cDetection of relationships between concepts
- dIndexing of exact search terms
- eTemporal classification of the source
Storage
All components are stored in the local knowledge database — separated by knowledge space.
Immediately searchable
Right after upload, the knowledge is available to the practice team for queries.
Two AI systems.
Continuous availability.
Two independent knowledge systems work in parallel. Both are populated simultaneously on upload — an outage stays invisible to the practice team.
Fast, resource-efficient retrieval with clear filtering options. Guarantees availability even under high load.
- Fast semantic vector search
- Exact filtering by document type
- Direct mapping to text segments
- Fallback during maintenance of the second system
Full multi-strategy retrieval with all eight capabilities — the qualitatively stronger search layer.
- Multi-strategy retrieval with four parallel paths
- Automatic extraction of entities and relationships
- Cross-encoder reranking after result fusion
- Insight derivation from existing knowledge
Both systems are populated in parallel on upload. If one fails, the other takes over — without data loss and without disruption to the practice.
Patient context stays
inside the practice.
Local processing
Uploaded documents and patient context remain on the AI server in the practice. There is no transfer to external providers.
Local language models
All language and embedding models run on the practice server. No API call to cloud AI services takes place — at any point in time.
GDPR by architecture
Data protection is not retrofitted but part of the architecture. Patient data does not leave the practice network.
Traceable sources
Every answer is linked to the document segments used. Transparency is a prerequisite for medical usability.
Isolated knowledge spaces
Knowledge spaces are technically separated. A query inside one module cannot inadvertently access content from another.
Practice data sovereignty
Uploaded documents, indexes and derived insights can be exported or deleted by the practice at any time — without involving third parties.
Knowledge management
for your pilot practice.
Pilot practices receive the full knowledge stack including onboarding and the initial population of the knowledge spaces.