Inspiration
We started from a simple observation: many real-world situations share the same underlying structure even when they are described differently. In law, contracts, CRM conversations, or other domains, the wording changes, but the roles, actions, and relationships often remain the same.
This led us to ask a simple question: What if we made that structure explicit before an LLM starts reasoning?
What we built
We built a Semantic Runtime Layer (SRL), an intermediate representation that transforms raw legal text into explicit semantic objects. Instead of sending text directly to a language model, SRL identifies elements such as Scene, Actor, Action and Object, preserving temporal context and traceable legal sources.
How we built it
- Civil Code PDF processing
- semantic normalization
- runtime construction
- temporal SQLite
- explainable output
- Codex + GPT-5.6
Challenges
- keeping legal meaning unchanged during normalization;
- preserving temporal correctness;
- separating the runtime representation from the LLM itself;
- presenting the idea clearly in a short demonstration.
What we learned
During the project we realized that the most valuable artifact was not another legal assistant, but the semantic runtime itself. Once the context becomes explicit, it can be inspected, validated, compared and reused before reasoning begins. Different texts can describe the same situation. SRL maps them into a shared semantic structure, making them comparable before reasoning begins.
Log in or sign up for Devpost to join the conversation.