Scaling experience across an AI data ecosystem.
Lantern turns messy fund data into signed-off, trustworthy numbers for private-capital teams. As product design lead I drive experience across several AI products - and I built Lux, the design system that holds them together.
Private capital runs on data nobody fully trusts.
Every quarter, fund managers receive waves of data from fund admins - valuations, returns, transactions - that has to be reconciled, validated and signed off before reporting. It arrives messy, late and inconsistent, and a wrong number is a real liability.
Lantern is the AI layer that catches the discrepancies and walks a team to a clean, signed-off quarter. The company grew into several products quickly - and that growth is exactly where the experience started to fragment.
Several products, several languages, one anxious user.
The work was not one screen - it was making a growing family of AI products feel like one trustworthy system to a finance professional who cannot afford to get a number wrong.
Each product spoke its own dialect
Patterns, components and data treatments drifted apart product to product - multiplying QA and quietly eroding trust.
AI numbers have to be auditable
A GP signs their name to these figures. The system has to show its working - the why behind every flagged difference, not just the what.
Calm under a flood of data
Long review sessions, dense tables, high stakes. The interface has to stay legible and unhurried when the data is anything but.
From feature execution to ecosystem leadership.
I lead product experience across Lantern’s AI products and own Lux, the design system underneath them. Day to day that means setting interaction and data-visualisation patterns, holding the accessibility bar, partnering with founders and engineering, and leading a team of four designers - shifting them from shipping features one by one to designing a coherent ecosystem.
Audit the sprawl, then build the spine.
Map the people and the sprawl
I audited the experience across products and grounded it in the people who live in it - the client-focused fund leader who answers to investors, and the quality-focused accountant who has to defend every figure. Their jobs set the bar: nothing ships unless it makes a number easier to trust.
One system, many products
I defined Lux: shared foundations, components and a data-quality language every product could speak. I mapped the information architecture across the ecosystem so a user moves between products without re-learning anything - and so explainability is a first-class pattern, not a bolt-on.
Mapping the whole quarter-end journey.
Before designing a single screen, I mapped the GP's quarter-end end to end - every step, action, question and pain point from management accounts arriving, through review and validation, to the handoff to LPs. It's the map that decided what Assure had to fix first.
The choices that made the ecosystem cohere.
A handful of decisions did the heavy lifting - each trading short-term speed for a system that scales and a number a GP can sign.
Invest in Lux, not per-product styling
One system instead of three dialects. Slower to start, far cheaper as the product count grows.
Make the AI auditable
Every flagged difference exposes its reasoning and a trail. Trust comes from showing the working, not asserting the answer.
A shared data-quality language
Data quality %, passed / failed, signed-off differences and missing - the same vocabulary in every product.
Calm colour under load
Emerald for trust and action; amber and red reserved for genuine attention. Never decorative, never noisy.
Decisions happen in context
Threaded change requests keep the discussion, status and history attached to the number they concern.
Lead the system, not the features
Designing the conditions and patterns so every product team ships consistently - without a bottleneck.
The thinking behind the first section: most data tools open with a wall of figures. This one opens with a sentence - what needs validating and why - so a fund manager knows exactly what they're looking at before they read a single number.
It opens with a sentence - what has been flagged and what to do - before a single number.
A running 0/62 differences-resolved ring keeps the size of the job, and how far through you are, visible the whole time.
The two things that erode trust - 39 differences and 23 missing - are surfaced and colour-coded up top.
Each client’s data-quality score is tracked over time, so a dip shows up long before quarter-end.
Data quality, turned into reassurance.
Assure is the data-quality product inside Lantern. It catches differences and missing data, then walks a team from a flagged client all the way down to the single transaction behind a number - three screens, one continuous thread.
Every client and fund, ranked by data-quality score - so the worst issues surface first.
Open a client: differences, missing data and a resolution-progress ring in one view.
Drill into a single metric - the test, the difference, and the transactions behind it.
Lux 4.0 was about doing less.
The colour and type were already settled - 4.0 was not a re-skin. It was a simplification pass: collapsing variants, merging separate screens into single adaptive components, and flattening the navigation, so a growing suite stays easy to learn.
A family of products that feels like one.
Lux now unifies Lantern’s AI products - shared components, a common data-quality language and explainability built into every signed-off number. Just as importantly, the design team shifted from executing features one at a time to leading a coherent ecosystem, with accessibility held as a baseline rather than an afterthought.
Leading a system is really about designing the conditions for other people to do their best work. The further I get from pushing pixels myself, the more leverage the patterns I set carry across every product.