January — October 2023
Raylu
Can a Figma file close a $4M round?
Two founders, a deep product, no artifact
Raylu was building UI for ML pipeline observability. The founders had product depth and technical clarity. What they didn't have was a single thing an investor could open.
It started as a wireframe pass for a pitch deck and became a six-month engagement: brand, a Figma component system, and clickable prototypes for fundraising and early-adopter conversations — everything between a founder sketch and a React-ready component library.
The bet was the success metric. Pre-PMF prototyping isn't shipped product. The brief was make the product readable enough to fund and build. Polish costs founder burn rate; clarity of concept doesn't. That trade-off shaped every decision.
The Figma file is the deck. The deliverable is alignment, not polish.
The gap was alignment
Sketches and Zoom calls produce ideas. They don't produce shared context across founders, investors, and an outside designer. Without a designed artifact, every pitch re-derived the product from a whiteboard.
So the first deliverable wasn't a screen. It was the Figma file as the alignment artifact — a workspace stakeholders could navigate without me in the room, with linear sections for brand, competitive analysis, strategy, and the product itself. Founders could self-serve pitch updates. I stopped being the bottleneck for every external conversation.
You just got the link
This is the file an investor opened cold. Scroll it the way they did — top to bottom, one page at a time. Brand. Competitive analysis. Strategy. Wireframes. Hi-fi. Each section dates itself, names itself, then fills in.
01Brand
First palette explorations, the whale mascot, and the headline thesis: “tools to make ML safe.”
- palette
- whale
- tone-of-voice
The whale mascot — picked because it reads friendly at any scale.
02Competitive analysis
Five sibling ML-tooling products mapped on a 2×2 of fidelity vs scope.
- Snorkel
- W&B
- Roboflow
- DataRobot
- Galileo
Kept the work honest — no surface someone else already shipped.
03Strategy / thesis
Goal: provide the tools to make ML safe. Deterministic vs unbounded — we built for the unbounded case.
- observability
- pipelines, not models
- human-rules over
Pipelines, not single models. Human rules over the chain. That positioning carried into every screen.
04Wireframes
Founder sketches translated into low-fi frames. Two starting paths: talk to a chat bot, or boot a premade system.
- chat-first
- premade systems
- two-path entry
Founder sketch (mid-Zoom call, then traced in Figma the same afternoon). The annotation became the V0 nav model.
05Hi-fi prototype
Pipeline builder hi-fi — model-integration nav, label-errors workflow, UMAP slicing. Clickable for investor demos.
- pipeline UI
- UMAP slicing
- label errors
The hi-fi pipeline — model nav, label errors, UMAP slicing. The investor didn't just scroll past this screen. In a moment, you'll open it yourself.
Sketch in the morning, frame by the afternoon
Two of those sections — Wireframes and Hi-fi — came from this loop. Founder sketches arrived over Zoom; clickable frames went back the same day. Drag the handle: a napkin sketch on the left, a real Figma frame from the same build on the right. No intent lost between the sketch and the screen.
Hi-fi Figma frame · same afternoon Founder sketch · mid-Zoom call Left of the handle: a founder's sketch from a Zoom call. Right: a real, clickable Figma frame from the same build, hours later. The rail and the intent carried straight through.
ML primitives a PM could read
The hard part was never the ML — it was making a real ML primitive legible to a non-engineer mid-demo. Here's the UMAP slice surface, live. Drag a box around a cluster of embeddings and it resolves into a named slice. Dimensionality reduction made into something you can grab.
vehicle · 47 images Drag a box around the green cluster — or press Enter to slice the vehicle cluster.
That's the gesture an investor made mid-demo. Now you've made it too.
You just clicked into this screen. So did they.
What followed
That gesture is the one an investor made mid-demo. The file held up where it mattered — investor demos without a designer present, early-adopter conversations, a workspace an engineering team could read.
seed round closed (2023)
Conversion Capital and Unusual Ventures led, with angels including Arash Ferdowsi.
I can't speak to a causal chain between my work and the funding. Design at pre-PMF is one ingredient among many — the product team did the long work, and the round closed under conditions that had moved well past anything the 2023 file specified. What I can speak to is what landed in the right hands: a navigable file investors could read cold, a brand the team could pitch from, and component patterns the engineering team picked up directly when they started.
The product has pivoted significantly since the 2023 work documented here — it's now an AI sourcing platform for private-market investors — and raised an $8M Series A in late 2025.
What generalizes
What I'd carry into any early-stage engagement — one principle, and one number.