I designed an LLM-assisted workspace that simplified a complex, fragmented strategy workflow and made it easier for quant strategy builders to monetize their work.
*
Original Workspace Problems:
- The workflow was fragmented and didn't match how builders work, making the platform harder to adopt
- Marketplace listings didn’t convert well
Solutions:
- The builder workflow now lives in a single workspace
- Builders can create stronger listings with the help of the LLM with full context
Working with the founding engineer, I designed the first version of the builder workspace for quantitative strategy builders, from 0 to 1. The workspace included an LLM assistant, Jupyter-notebook coding section, and a right panel to submit strategy code for validation.
Validation was a necessary step if builders wanted to deploy their code or list their strategies on the marketplace. Due to initial tech constraints, validation lived on a separate page, not in the workspace.
We noticed that builder-created strategies were not performing well on the marketplace, despite these strategies showing strong investment returns. One key reason was that the listings builders created did not look professional or polished. More importantly, they did not contain information traders cared about, such as how a strategy actually trades and its risk assessment.
“No idea how this trades”
“Looks unreliable”
We wanted AI to help builders create their listings. Rather than adding a new AI feature on the listing creation page, I proposed that we move listing creation, and therefore validation, directly under each project in the workspace. The LLM chat already had full context of a strategy, so this would be the most token-efficient approach.
Moving validation inside the workspace would also let bot deployment happen there, meaning the entire workflow could live in one place. No more jumping between pages or context-switching. The CTO agreed to this direction, so we overhauled the right panel to support this change.
New validation right panel
Option to create listing with AI
AI suggesting the next step after listing is drafted
LLM-assisted listing creation - right panel expanded view
To reflect this change, I also redesigned the LLM chat logic. Previously, builders had to “submit” code for validation, which was a term that didn't match how builders think. 'Deploy,' instead, fits their mental model and workflow. Now, when the AI detects intent to deploy, it just prompts builders to validate their strategy code
The chat guided users to “submit” code for validation - an unfamiliar step for users.
The chat now prompts users to validate once it detects intent to deploy.