JuriSpark

Three-Pillar Solution for Legal Strategy, Simulation & Prediction

Comment

JuriSpark.com : Advanced AI-Powered Litigation Strategy Engine

Inspiration

We were a group of students from University of Cambridge inspired by Simmons & Simmons (a leading law firm) challenge to rethink how legal teams manage expert reports in complex litigation. With hundreds of pages across multiple expert submissions, legal professionals currently face a slow, manual process to identify what each expert says, where they agree or disagree, and how their views align with final case outcomes.

The firm’s forward-thinking approach, particularly through Simmons Wavelength and Percy AI, encouraged us to think bigger: not just about managing documents, but about actively supporting legal strategy with cutting-edge AI and simulation.

What it does

JuriSpark.com is a three-pillar AI system designed to support legal teams in high-stakes litigation by:

  1. Structuring case knowledge through a dynamic knowledge graph that extracts and maps relationships, flows, and timelines across case documents.
  2. Analyzing arguments using ML and generative AI to identify persuasive patterns, optimize wording, and surface strategic weaknesses.
  3. Simulating debates between AI agents that represent both parties and a judge-like decision model, providing probabilistic insights into argument success and identifying pressure points before trial.

It enables natural language queries like:

  • "Which experts disagree on loss causation?"
  • "What was Expert B’s position on governance failures?"
  • "Which arguments were persuasive to the tribunal?"

How we built it

We built JuriSpark over three phases:

  • Phase 1: Knowledge Graph Engine
    Using NLP and entity recognition, we structured a dynamic graph to map key experts, events, dates, case citations, organizations, regulations, concepts across the full case corpus, and identify the relationships from 3 dimensions (financial, physical and information flow).

  • Phase 2: Argument Analytics
    We created a sentence-level argument evaluator, scoring alignment from expert reports, merits with final tribunal outcomes. It shows how much a specific evidence is more likely to be taken into final awards. This component powered a generative module that improved legal phrasing and strengthen the arguments chain.

  • Phase 3: Simulated Debate Framework
    We built a multi-agent environment where the Claimant and Respondent AI agents argue against each other, judged by a Tribunal agent trained on legal reasoning patterns. We ran 100 simulations to derive winning probabilities and debate flow trajectories. The Junior lawyer could learn from the debate and also get helped by AI on complicated cases.

Challenges we ran into

  • Data disambiguation: Aligning entities and timelines across inconsistent expert reports was labor-intensive and error-prone.
  • Time limitations: Designing three full modules with interaction logic and ensuring model alignment to legal context within a hackathon schedule was a major constraint.
  • Grounding AI outputs: Ensuring that generative suggestions always tied back to real evidence required custom RAG workflows and rigorous validation.
  • Simulating legal logic: We had to go beyond generic LLM outputs and define a framework that reflects structured legal adjudication, not just language generation.

Accomplishments that we're proud of

  • Delivered an end-to-end, multi-layered AI solution that spans evidence structuring, argument scoring, and litigation simulation—all within 48 hours.
  • Successfully built a dynamic knowledge graph and ran AI-powered simulations based on real case documents.
  • Developed a modular, extensible architecture that could integrate with tools like Percy AI and scale for real-world law firm deployment.
  • Bridged technical, legal, and business perspectives into one cohesive strategy system.

What we learned

  • Legal strategy is not just about facts—it's about narratives, timing, and relationships.
  • Generative AI needs domain-specific grounding and structure to be truly helpful in law.
  • Collaboration between technical and legal minds can unlock powerful new capabilities in traditionally conservative industries.
  • Simulations aren’t just for prediction—they’re tools to foster confidence in decision-making.

What's next for JuriSpark.com

We’re excited to take JuriSpark.com further. Next steps include:

  • Expanding the simulation framework to support user-defined argument insertion and live strategy workshops.
  • Piloting JuriSpark.com on anonymized past cases with law firms for validation and refinement.
  • Exploring additional applications beyond litigation, including regulatory reviews and compliance investigations.

One More Thing:

  • Spinned out from the same model, we build a demo to specifically process legal documents. Check out the new demo at YouTube here.

JuriSpark.com aims to become a trusted AI co-counsel—one that empowers lawyers with insight, clarity, and strategic foresight.

Built With

Share this project:

Updates

Submission history