Vallor Internship — Shriya Chipde

SHRIYA CHIPDE

Designing AI for Procurement: My Spring Internship at Vallor

This case study explores how I redesigned core pages of Vallor's AI-powered procurement platform as the company's only product designer. The platform offers a powerful suite of AI capabilities, chart generation, contract redlining, automated playbooks, and conversational task execution, but its interface hadn't kept pace with what the underlying system could do, leaving users navigating advanced AI through pages that felt outdated and disconnected from the intelligence beneath them.

ROLE
Product Design Intern

DURATION
15 Weeks

TOOLS
Figma, Cursor, v0, Claude code

OVERVIEW
Vallor

Vallor is a seed stage AI company building tools for procurement teams; the people inside organizations responsible for sourcing vendors, managing contracts, and controlling spend. It's a function that touches every part of a business but rarely gets good software built around it.

The team is small: thirteen people, mostly engineers. When I joined as a product design intern, there was no designer driving the work forward. An established design system existed, but pages needed rethinking and features needed building. I worked most closely with the CEO and collaborated across engineering and product depending on what we were shipping.

THE PROBLEM
The Gap Between Capability and Experience

Vallor's AI can redline contracts, generate charts, set up automations, define playbooks, and carry out complex procurement tasks through a single conversational interface. The capability was already there. The interface hadn't completely caught up.

Three core pages had distinct UX problems. A major new feature was waiting to be built from scratch. And no one was translating what the AI could do into something a user could feel.

The question driving every decision: How might we make complex AI systems feel intuitive and approachable without requiring users to learn new interaction patterns?

THE ROLE
Design, Prototype, Ship

As the only product designer, I owned the full process: UX research, designing within the existing system, prototyping, and opening PRs to implement UI changes directly. Figma was home base. v0, Cursor, and Claude Code let me move from design to code and ship alongside engineering instead of handing off and waiting.

METHODOLY OVERVIEW
My Process and Tools Used

THE SOLUTION
The Three Pages That Needed the Most Attention

01. Charts Page

Visual polish and UX friction in how users edited generated charts.

THE SOLUTION
Rebuilt the page layout and reorganized the editing components so users could go from generated chart to finished chart in fewer steps. The chart canvas and its editing controls were reorganized around how users actually moved between them, not how the components had originally been grouped.

Design Iterations

Final Design

THE IMPACT
Editing a generated chart became a continuous flow, and the refined chart colors made the interface feel more cohesive and intuitive.

02. Redlining Page

Users couldn't tell at a glance what the AI had flagged, what it had changed, or what action to take next. A capability that should have been a headline feature was buried under an interface that couldn't keep up.

THE SOLUTION
A full page replacement, built from the smallest unit up:

Started with badges. The page leaned on them heavily, so getting the badge system right at the foundation would carry through every layer above it. I defined a full set of badge styles tied to redlining states, source types, and AI activity. I also focused heavily on color contrast, lighter color palettes required more careful consideration to maintain visual subtlety while still meeting WCAG accessibility standards.

Redesigned the card stacks for the Playbook and Redlines tabs. The key move: users can now act directly from a collapsed card without opening it. Small interaction change, significant workflow gain — users stopped having to drill into a card just to dismiss or accept a single suggestion.

Playbook card stack

Redlines card stack

Rebuilt the open card view to give each redline a clear visual hierarchy: what changed, why the AI flagged it, and what the user can do.

Redesigned Agent Activity so users could finally see, in one place, what the AI was doing on their behalf.

Opened PRs to implement components from the redesign directly — pushing UI changes into the codebase alongside engineering rather than handing off and waiting.

THE IMPACT
A redlined contract now reads at a glance. The page shipped within a single sprint, proof that design could move at engineering speed without sacrificing quality, and the moment the team's relationship to design shifted.

03. Dashboard Design

The original dashboard was a chat input and some charts. We wanted it to become the operational center of the product, a place where users could chat with the AI, track tasks, see team activity, and access artifacts in one view.

The hardest sub-problem: how do you let one chat input do five different things — redline, save, compare, task, and chatmwithout making the user think about which mode they're in?

THE SOLUTION

Chat Input:
Designed an intent-detection pattern: the AI auto-detects what the user is trying to do based on the file type and content they bring in. The user confirms by continuing, or switches modes with a tap. No mode-switching menus. No upfront decisions. The complexity moves from the user's head to the system.

The dashboard around it: Extended the redesign into a full layout that surfaces tasks, recent team activity, and saved artifacts alongside the chat. I sketched multiple wireframe directions, gathered feedback from the CEO and engineering, and converged on a layout that put action — not conversation alone — at the center of the page.

Ideation: Layout Wireframes

After feedback on the layout options, we landed on this layout:

THE IMPACT
The dashboard went from a single-purpose chat surface to a multi-purpose workspace. Five distinct user actions now flow through one input without users having to think about it. The page is still in active development.

04. Tasks Feature Design

There was no Tasks feature when I joined. The Dashboard redesign needed it, and procurement workflows demanded it — but it had to be built from a blank file.

THE SOLUTION
Research first. Ran a competitive analysis across adjacent task-management products, reviewed user sessions to understand how Vallor users were using the product and synthesized the gaps into a feature brief.

Competitive Analysis

Architecture next. Presented findings to the team, and together we defined the feature's structure: how tasks are created, where they live, and how they surface on the Dashboard.

Then design. Iterated through multiple directions before landing on a solution that fit Vallor's existing design language and supported the workflows users were already trying to build.

This feature is currentling being developed and the tasks card component from the dashboard pulls tasks from this page.

Reflecting on the past 3 months:

IMPACT AT A GLANCE
Establishing Design at Vallor

Shipped a full Redlining page redesign within a single sprint, establishing that design could move at engineering speed.

Designed a new chat input pattern that lets one input handle five distinct user actions through AI intent detection.
Took the Dashboard from a chat-and-charts page to a multi-purpose workspace, surfacing tasks, activity, and artifacts in a single view.

Built the Tasks feature from, research, architecture, and design

Opened PRs to ship UI changes directly, closing the design-to-implementation gap and letting design move at the same speed as engineering.

Established a design presence that hadn't existed before. Patterns, processes, and a way of working the team can build on.

TAKEAWAYS
What Stuck With Me

Trust came from being transparent and involved in the work itself. Sharing rough work early built credibility, but opening PRs to implement my own designs changed the relationship entirely. Design stopped being treated as a separate phase and became part of the build process itself.

The designer's most valuable skill on a small team is translation, converting user needs into something engineers can build and stakeholders can trust.

When the AI is the product, the interface is the trust layer. Every design decision was really a decision about how much the user could rely on what they were seeing.

My workspace in our virtual office :)

Email LinkedIn