Contract Tree | I am Amrit

CASE STUDY -> SYSTEM DESIGN

CASE STUDY

Which amendment applies to this SOW?"

Users couldn't answer !

From storage hierarchy → relationship-driven system.

ROLE

Product Designer

CONTEXT

Enterprise CLM platform

FOCUS

System design, AI enablement

YEAR

2026

My role.
As an AI-native Product Designer, I owned the redesign of the contract data model that Contract Store Intelligence runs on. I treated this not as a UI cleanup but as the foundation that makes contract AI trustworthy. I reframed 10 user-reported problems (PERs) into one insight the system had to represent legal relationships, not folders and made the core architectural call to model contracts as entities: MSA → SOW → Amendment.

-40%

REVIEW TIME

Efficiency gained by removing hierarchy reconstruction tasks.

100%

DATA INTEGRITY

Elimination of conflicting "current" versions in the system.

85%

AI ACCURACY

Context-aware extraction with minimal human correction.

In the existing system

Contracts as Folders

Users navigated through a nested directory of PDF files. To understand the current state of an agreement, a lawyer had to manually open 5-10 different files and mentally reconcile the changes.

The system was lying
about structure

It showed structure as folders.
But contracts don’t work like folder - Contracts have historically been treated as static files stored in folders. This "flat" storage model fails to represent how legal obligations actually function: as a living, interconnected web where one document can modify, terminate, or extend another.

THE CHALLENGE

"This looked like a search problem.

So we tried everything -> filters, OCR, indexing.
Nothing worked. Because users weren’t searching. They were reconstructing meaning

"The core problem wasn't a lack of search or better OCR, it was the fundamental absence of a structural data model that mirrors legal intent."

Contracts have authority, scope, and evolution. None of these were modelled. So users built them manually — in file names, Excel sheets, Confluence pages — outside the system entirely.

The Decision Moment

Optimize the folder browser —
or rebuild the data model from scratch?

We chose to rebuild. A harder technical lift but the only foundation on which the system could match how law actually works, and the only infrastructure on which reliable AI could be built

Is Relationship-Driven Contract Design the Future?

View Full Report

THE CONCEPTUAL SHIFT

From storing contracts →
to representing relationships

A contract isn't a file. It's a system made of three things: Authority (the master agreement), Relationships (SOWs and amendments), and Evolution (time and status).

Structure

Defines the hierarchy of authority. Every document knows its 'parent' and its scope of influence.

Relationships

The connecting lines that show hierarchy and precedence across the entire portfolio.

Documents

The physical evidence. The file becomes an attribute of the relationship, not the container of it.

A Clearer View

Visible Relationships

Instead of digging through PDF pages, legal counsel can now see a tree of influence. Which amendment changed the liability cap? It's now a primary visual signal.

Authority Clarity

The system surfaces the active clause automatically and suppresses superseded text, freeing reviewers from manual reconciliation while keeping an audit-ready history.

Three systems, one coherent model

A relationship graph for navigation. Persona-specific workspaces that surface only what's relevant to each role. A contract overview that replaces the folder browser entirely. All three powered by the same underlying relationship data model.

Vibe coded html concept to showcase the idea

Hierarchy Visibility

A system-driven sidebar that shows the logical parent-child structure of an agreement across multiple years and entities.

"You can't layer AI on top of a broken structure. Structure defines what AI can understand."

By providing a structured data model, we eliminated the hallucinations common when LLMs try to parse flat folders. The AI now has a 'map' to follow, ensuring it pulls data from the correct, most recent version of a legal clause.