Technology
Symbolic AI. Determinism. Transparency.
Clausa combines neural language models with symbolic rule engines. The result: contract review that does not interpolate — it verifies.
What is Neuro-Symbolic AI?
Neural models (LLMs) understand language, but they interpolate. Symbolic systems follow formal rules but cannot process natural language. Neuro-Symbolic AI combines both: the language understanding of the LLM with the precision of the rule engine.
Why determinism matters in legal analysis
- Law is structure — not probability. §536 BGB and BGH clause rules follow facts-subsumption-legal consequence. LLMs do not know the difference.
- LLMs hallucinate legal provisions. Symbolic rules do not: they affirm or deny based on defined conditions — reproducibly.
- Auditability: Every decision by the symbol layer is traceable — not as a confidence score, but as documented rule application.
The Clausa Architecture
- LLM Layer: Understands contract text, recognises clause types, extracts parties, deadlines and conditions in natural language.
- Symbol Layer: Checks extracted elements against legal rule sets — BGB tenancy law, KSchG, TzBfG, BGH/BAG case law.
- Output: Structured review result with rule reference, paragraph source and human escalation flag for edge cases.
LLM-only vs. Neuro-Symbolic
| Feature | LLM-only (ChatGPT) | Neuro-Symbolic (Clausa) |
|---|---|---|
| Paragraph accuracy | Interpolated | Deterministic |
| Reproducibility | Varies per query | Identical for same input |
| Clause validation | Probabilistic summary | Rule-based checking |
| Source transparency | Not guaranteed | Paragraph reference per output |
| Hallucination risk | High for paragraph details | Minimised via symbol gate |
Frequently Asked Questions
Is Clausa an LLM wrapper?
No. Clausa uses LLMs as one layer — for language understanding and clause extraction. The review logic sits in a separate symbolic rule engine that works deterministically.
What is the difference between neuro-symbolic and RAG?
RAG (Retrieval-Augmented Generation) supplements an LLM with context documents. That improves answer quality but does not solve the interpolation problem. Clausa's symbol layer checks against formal rule sets — independently of the LLM output.
Can the symbolic layer respond to new case law?
Yes. Rule sets are maintained by the Clausa team and updated on relevant BGH/BAG decisions. The update process is documented and versioned.
For which contract types does neuro-symbolic review work?
Currently for lease agreements (BGB tenancy law, BetrKV, BGH case law) and employment contracts (KSchG, TzBfG, BAG case law). Law firm contracts follow in Phase 3.
Test Clausa in your sector
We are selecting 10 pilot partners. Pre-launch conditions.
Join the Waitlist