Chisquares Platform

The context

Chisquares is a US-based company building AI-powered analytics tools for academic and professional researchers. The platform aggregates research data, surfaces statistical patterns, and helps researchers synthesise findings across large bodies of literature and data.

I joined as the Lead Product Designer in March 2023 and was responsible for the entire product design function — from user research through to production handoff — across a two-year engagement. The team was fully remote; I worked from Lagos while collaborating with an engineering team spread across the US and Eastern Europe.

The problem

When I came on board, the product had shipped its initial version but was struggling to retain users past the first two weeks. The core AI features were genuinely powerful — the problem was that the interface made them feel inaccessible. Researchers didn't trust the outputs, couldn't build a mental model of the platform, and couldn't easily move between the tools.

Three concrete problems to solve:

• No coherent information architecture — features were organised around engineering modules, not research workflows
• AI outputs were opaque: summaries and pattern-detection results had no provenance, so users couldn't evaluate or cite them
• No design system — each feature had been built independently, creating a visual and interaction language that was inconsistent and expensive to maintain

Discovery

I started with a structured research sprint: 12 moderated sessions with researchers across academia and corporate R&D, plus analysis of session recordings and drop-off points in the onboarding funnel. The picture was consistent: users understood the value proposition immediately, got stuck in the first session, and didn't come back.

The drop-off wasn't a feature problem — it was an orientation problem. Users couldn't figure out where to start, and the AI outputs gave them nothing to hold onto. "It summarised my 200 papers — but how? What did it weight? Can I actually use this in a paper?"

Rebuilding the design system

Before touching product flows, I built the design system. I audited every screen, catalogued components and patterns, and documented the inconsistencies. Then I built a new system in Figma: semantic colour tokens, a type scale, a grid, and a base component library with documented usage rules.

The system shipped incrementally alongside product work — not as a separate "DS project" — so engineering was adopting it in real time. Within three months, the front-end codebase had a shared component library that matched the Figma specs 1:1.

Restructuring around research workflows

The new IA was built around the actual sequence of a research project: collect → organise → analyse → synthesise → export. Features were reorganised and relabelled to match this language, with a persistent "project" concept as the unifying container. Users could now orient themselves immediately: "I'm in the analysis phase of this project."

Making AI outputs accountable

Every AI-generated output — summaries, pattern detections, citations — was redesigned to include a transparent source trail. Users could expand any AI output to see the specific papers and passages it was derived from. A confidence indicator showed the signal strength. Everything was directly citable.

This change required close collaboration with the ML team to expose the right metadata from the model outputs — it was as much an engineering design problem as a UX one. The result was that the AI became a tool researchers could trust and use in published work, rather than a black box they had to verify independently.

What I learned

AI-powered products live or die on explainability. Every interaction where the system produces an output is an opportunity to either earn or lose trust. Design for researchers specifically taught me that the bar for explainability is very high: they need to be able to stake professional credibility on what the tool produces. That constraint made the design better.

The design system work was also foundational in a way I didn't fully appreciate until afterwards. The productivity gains from having a shared language between design and engineering compounded across the two-year engagement — by the end, we were shipping new features in days, not weeks, because the infrastructure was solid.