Pulse
An AI data room for founders raising money
Pulse is a startup building an AI-powered data room for founders raising money. The product has two sides a founder side for sharing documents and managing an investor pipeline, and an investor side where investors review shared materials and run diligence with Perry, Pulse's AI assistant.
Pulse's team asked me to build a design system from scratch, redesign both sides of the product, and rebuild the landing page.
The team
Worked with Christian (co-founder) and Ipalibo (co-founder and engineer), using Figma, Claude Code MCP, React, and Notion.
Findings
What I found when I joined.
No design system. No tokens with roles. No documented components. The product worked, but it didn't have a foundation to sit on.
The features were live and the flows worked. The opportunity was structural: clarify the hierarchy, align navigation across surfaces, and surface what mattered most.
Pulse didn't need rebuilding. It needed a foundation, and a structural, intuitive navigation and layout.
The approach
How I approached the work.
Compress the audit into the doing.
Reports about inconsistency rarely move teams. I rebuilt components into Figma as I encountered them in the code. The audit became the system.
Tokens before components.
Every value as a Figma variable, matching the names in code. No hardcoded hex anywhere. Every decision after that became a token conversation, not a hex one.
Composition over duplication.
One component, structural variation in the wrapper. If I reached for a variant, I asked first whether the change could move outside the component.
Jobs over features.
Every layout decision started from “what is the user trying to answer right now?” The questions changed by surface. The framing didn't.
03 — Foundation
Building the design system.
Step 01
Joined the repo branch and used the Claude Code MCP plugin to bring the current designs into my Figma file.
Step 02
Rebuilt the token layer with semantic roles for color, size, and typography, all meeting WCAG AA.
Step 03
Documented the core components, with naming consistent from Figma to code.
Redesigns
Redesigning the product.
Two surfaces. Different jobs, different problems.
Founder side.
The founder side worked but didn't feel like a fundraising tool. Contacts was a flat list of viewers a log, not a CRM. I framed the redesign around the five questions a founder asks when they open Pulse: what happened, who is hot, who needs a follow-up, which shares are at risk, how is the raise going.
The redesign repositioned Contacts as the investor pipeline. Hot / Warm / Cold grouping replaced the flat list. Each card carried an interpreted signal “4/4 docs · Engaged” not a raw count. Contact detail moved to two tabs (Overview, Shared) with Activity as a collapsible side panel.
The structural piece was the WorkspaceLayout master one Figma component with three Slots (CollectionPanel, DetailPanel, ContextPanel) handling every founder view: Drive, Contacts, Links. Composition over duplication, made real.
Investor side.
The investor side had a worse version of the same problem. It opened to a dead-end screen: “Select a Data Room.” Perry the AI differentiator of the product was hidden behind a click. The diligence checklist disappeared the moment a document was opened.
I started with the obvious move: reuse the three-panel WorkspaceLayout from the founder side. Data rooms on the left, deal home in the middle, Perry on the right as a collapsible panel. It solved the dead-end. It looked competent. It felt incomplete.
The investor isn't entering Pulse to manage a workspace. They're entering to read documents and understand them with Perry's help. Perry isn't a side panel. Perry is the work.
So the layout changed:
State A Default
No document open. Perry occupies the entire canvas with “What can I help with?” and an input pinned at the bottom. Sidebar carries Shared Data and Workspace Materials. Data rooms sit as a rail on the far left.
State B Document open
The canvas splits. Document on the left, Perry on the right as a resizable panel. The input placeholder shifts from “Message Perry…” to “Ask about this document…”
Two decisions worth calling out:
Sidebar collapse is manual, not automatic.
An early version auto-collapsed the sidebar when a document opened. It felt disorienting the app was deciding things for the user. A chevron made it explicit. Auto-behaviors reduce intuitiveness; explicit controls preserve trust.
Citations are bidirectional.
Perry's responses carry pill citations like [1] p.4. Clicking jumps to the cited page with a highlight. This is what makes the split layout earn its keep chat and document talking to each other, not just sitting next to each other.
Workspace Materials in the sidebar holds checklists and tables. First instinct: differentiate them with cards and progress bars in the sidebar. I pulled back. Sidebars carry navigation, not content. A unified list, equal rows, differentiated by icon shape and color. Progress bars and full checklists live in the main area.
"The sidebar's job is to get you somewhere. The main area's job is to show you something."
Landing page
Rebuilding the landing page.
With the product locked, the landing page came last. The hero went through a few drafts before landing on “Smart data rooms for fundraising.” Five words. One thing.
The structural change was reframing features around jobs the founder is trying to get done, not features Pulse ships.
“AI Search” became “Let investors find answers themselves.”
“Access Control” became “Share with the right people, instantly.”
Same content. Different reaction.
Perry got its own full-bleed section after the features grid the differentiator deserved the spotlight, not a column. Six feature visuals down the page, all using the same fictional cast (Acme Inc, Maria Alvarez at Accel, James Tan at Index Ventures, Lina Reyes at Sequoia Scout, David Park at Founders Fund) for visual continuity.
Reflection
What I learned.
Lesson 01 The work made the audit unnecessary.
By the time I had a system, I'd already answered the audit's questions by rebuilding. Reports about inconsistency rarely move teams. Design systems that absorb the inconsistency do.
Lesson 02 The fix was structural, not functional.
Pulse's features were right. The information architecture wasn't. Once the foundation existed and the hierarchy was fixed, the product's actual strengths could surface.
Lesson 03 Auto-behaviors get punished by users.
The investor sidebar auto-collapse was clever and felt wrong the moment a real person used it. Explicit controls survive contact with users in a way that helpful magic doesn't.
Pulse · 2026 Designed by Shay Khezr. The live product is up at usepulse.co.
Visit usepulse.co