TrainLoop

Building TrainLoop's brand and product around gradient descent

Company: TrainLoop AI (YC W25)
Year: 2025 - 26
Industry: AI
Contributors: Jackson Stokes, Mason Pierce

Results

Seed from Y Combinator: $500k
From no defined ICP to multi-million-dollar partnerships: < 1 year
Design system across product, site and docs: 1

Context

Enterprises didn't trust a young AI startup yet

TrainLoop fine-tunes LLMs on a company's own policies, docs and workflows. Talking to enterprise buyers, one worry kept coming up: hallucination, with no way to see why a model answered as it did. As founding designer I led brand and product end to end, so trust had to show up in both.

01Platform

Inside the platform

Projects, execution history, evaluation and discussion in one place, from the project overview down to a single decision node.

Why: Buyers couldn't see why a model answered as it did, so every step of a workflow is a task, verifier or decision node you can open and inspect.

TrainLoop: Project dashboard: active projects, execution history, eval metrics, libraries and docs. — 1.1Project dashboard: active projects, execution history, eval metrics, libraries and docs.

TrainLoop: The orchestration builder. Reasoning workflows laid out as execution graphs; any task, verifier or decision can be inspected. — 1.2The orchestration builder. Reasoning workflows laid out as execution graphs; any task, verifier or decision can be inspected.

TrainLoop: Messaging, where researchers, engineers and evaluators talk through experiments and share assets. — 1.3Messaging, where researchers, engineers and evaluators talk through experiments and share assets.

TrainLoop: Contributor profiles with projects, activity, reputation and org context. — 1.4Contributor profiles with projects, activity, reputation and org context.

02Branding

The identity

One idea, refinement over time, carried across the mark, colour and design system.

Why: Gradient descent comes straight from the world our buyers work in, so the mark is a dot matrix that converges like a model learning.

TrainLoop: Early territories from science, computation and knowledge systems. — 2.1Early territories from science, computation and knowledge systems.

TrainLoop: The final mark. Dots shift in density and brightness, like a model converging over training steps. — 2.2The final mark. Dots shift in density and brightness, like a model converging over training steps.

2.3The design system that grew out of it, used across product, marketing, docs and brand comms.

03Website

Website and case studies

Editorial, not loud. It leads with what TrainLoop actually did for teams.

Why: Buyers were working out whether they could rely on us, so the site leans on research credibility and operational transparency.

TrainLoop: Editorial landing page, built on clarity and technical depth. — 3.1Editorial landing page, built on clarity and technical depth.

TrainLoop: Enterprise case studies focused on measurable gains in agent performance and operational reliability. — 3.2Enterprise case studies focused on measurable gains in agent performance and operational reliability.

Outcome

What changed

$500k seed from Y Combinator, plus multiple undisclosed rounds.
From no defined ICP to multi-million-dollar partnerships within a year.
Orchestration became inspectable reasoning workflows, with more visibility into execution, verification and model performance.
Collaboration, evaluation and deployment in one place.
One design system across product and marketing, with reusable UI patterns for future AI infrastructure features.
A brand that reads as scientifically rigorous and enterprise-ready.

“These are SOSO cool, I’m looking through now but honestly I love all of them so it will be tough to choose”

JacksonTrainLoop AI

Hiring for product design, or building something new? Send the context and we can talk it through.

Book a 30-min call (opens in new tab)

NextRekonsile Cutting seller reconciliation time by 60%