Detect contradictions in legal documents

Statements are negotiable, but integrity is not.

Corporate reporting is under growing scrutiny. As economic volatility rises, regulators demand more detailed disclosures — and even a single number can carry legal liability. Yet verification still relies on fragmented manual checks across teams and document versions, an approach that struggles as reports grow longer and more complex. Atticus was designed to modernise this process, making verification collaborative and secure by default, but one critical gap remained: maintaining consistency as documents evolve.

As LLMs began proving their potential in the legal domain, an opportunity emerged to build a 0→1 generative AI capability with no existing playbook.

This project allowed me to demonstrate how I drove alignment and shaped product direction on a three-year challenge in under five months, through continuous discovery, rapid experimentation, and strategic framing.

Key decisions I drove:

• Establishing consistent content frameworks for non-deterministic AI outputs

• Building a human-centered evaluation model to scale product quality

• Stress-testing AI workflows under real-world constraints