2018, new york city → stockholm → london → palo alto → washington
I extensively travelled for 4 months as a product design intern at Palantir, working closely with clients and coworkers overseas to quickly iterate and ship features on Dossier and Gaia, our collaborative report-writing and map modules, respectively.
*Most of my work is under an NDA. If you want to learn more, shoot me a message.
I worked on Palantir Gotham (PG), which houses a suite of analytic products that are focused on a bottom-up approach to answering big data questions — making PG a popular platform for investigatory workflows. The different features and modules included in PG output their own analytical pieces, such as objects, maps, tables, documents, and graphs.
Most of our PG clients delivered formal reports, which were heavily audited before being published. Dossier aimed to bridge this gap between building analytical pieces in PG and augmenting them with written, actionable reports.
I worked with a team comprised of a tech lead, two developers, and my mentor and product manager, part-time.
Dossier has a homepage which includes all the dossiers available for the user's investigation, along with its metadata (such as document labels—which I also designed, user tested, and shipped with one of the developers on my team!)
The actual documents are called dossiers, complete with classification and permission settings, which dictate who and what pieces can enter the dossier.
Collaborating closely with our early adopters, I focused on designing various low-hanging fruit features for easier adoption of Dossier in their day-to-day workflows. This included focusing on automating tasks such as citations.
“We're looking for automation: getting properties and their sources, appendices—and whatever else is the same every single time—into Word.”
As formal report writers, our clients needed a way to easily source all their pieces in PG—and sometimes all the way to the property- and value- level of a single object. These reports could include up to more than 1000 citations, which they manually format by copy-pasting.
The end result was an early adopter group going from being testers of our product to becoming champions of Dossier in their organization. With formatting and automation tasks out of the way, users were able to seamlessly do their report-writing in Dossier and minimally format it in Word for final delivery.
To fully make Dossier a canvas of analytical pieces and hypotheses, we needed to improve its ability to tie in together the different outputs of features and modules within PG.
“Collaboration can be encouraged by Dossier, and I'm a big believer in turning PG into an institutional knowledge base.”
This problem obviously encompassed cross-platform discoverability, but for easier developer hand-off and adoption from stakeholders, I decided to scope the project into various versions. I started with improving prompts and discoverability within a dossier.
To consolidate all these workflows and shortcuts, I created a flowchart for all possible piece insertions and the different ways they can be inserted into a dossier.
The result was,
Due to more urgent features (such as sourcing and citation), this work was postponed. Nonetheless this was my first project at Palantir, and I'm satisfied to have decomposed a systematic design problem so early on.
Lastly, I worked on Gaia (maps) for my last few weeks at Palantir, where I researched and created early designs for extending our map visualization features.
Working on another large analytical artifact, I improved maps by putting together new map visualizations for our Gaia module.
Working with various clients who heavily use our map visualization features, an important lesson I learned were the two main uses of map visualizations — presentation and analysis.
I focused on presentation-type map visualizations, dealing with multiple colours and varying intensities. The biggest challenge with controlling these visualizations is knowing what the variables are, and how they affect your visualization.
To start off, these visualizations often compare densities of something. A frequency. A point's property value. When densities are in play, users would want to tend to focus on clusters that show extreme low or high. But there is a cost to beauty. These polished gradients and vibrant clusters are controlled by variables that change based on statistical distribution equations—which smooth your gradients and weigh your clusters accordingly.
Another improvement made was the control of colour. Colour stops are an important feature when fine-tuning the semantics of map visualization. This also allowed for the user to easily map what values the colours corresponded to without having to provide multiple fall-off points, decluttering the UI.
Moreover, I researched on more accessible palettes (read: I will never stop playing with and raving about Cubehelix) to provide out of the box for our users. My palettes of choice became Cubehelix for print, Viridis for digital, and ColorBrewery for choropleths (especially for qualitiative and divergent sets).
This was a very fun project to end off my internship, and having colocated with a new team in Washington, D.C. It was exciting to come back to my data visualization roots, combining statistics and visual design knowledge in a product. Moreover, I was able to experience more facets of the PG product, and Palantir as an organization.
Ashley Einspahr, Andy Elder, Alexis Le Dantec, Ryan Beiermeister, Matt Bango, and the PG designers for your guidance. Rhys Brett-Bowen and Paul Thoren for the great jams and your developer chops.
And Kevin Ng, my mentor, work best friend, and introversion coach.