Anup ShresthaCase study

DVLP Studio

DVLP is a platform where pharmaceutical teams plan drug programs, assemble the evidence, and eventually trade the assets themselves. Across 2025 and 2026 I redesigned it end to end onto a lab-paper system: a warm bone canvas, a single oxidized orange, one grotesk doing all the talking. Nine surfaces of dense scientific data, made calm enough to work in all day.

Role: Design engineer
Years: 2025–26
Stack: Next.js · Tailwind · Design system
Scope: 9 product surfaces

DVLP Studio asset home on a bone-paper canvas: sidebar navigation, phase IIa progress bar, a recent activity feed of verified documents, and a Drug Context Information Inquiry panel on the right showing an AI conclusion with supporting sources — Fig. 01The asset home. Phase progress and recent activity on the left; on the right, the drug context inquiry panel answers questions from the asset's own records and cites its sources.

One surface for the science

Every asset gets a home that answers three questions the moment it loads: where the program stands, what needs attention next, and what changed while you were away. The AI panel sits alongside rather than in a separate chat product, because in this domain an answer without citations is worthless.

The development plan lives one click deeper. Clinical, commercial, non-clinical evaluation, pharm ops, CMC, and regulatory workstreams share a single board, cut by stage gates. Each deliverable carries its evidence status, so a program lead can spot the thin spots without opening a single document.

Development plan explorer showing six swimlanes for Clinical, Commercial, non-clinical drug evaluation, Pharm Ops, CMC process development, and Regulatory, with stage gate markers and deliverable cards colored by verification status — Fig. 02The plan explorer. Six workstreams against shared stage gates, every deliverable showing whether its evidence is verified, partial, or missing.

Data you can walk through

A drug asset is thousands of files and the entities they describe. I built two ways through it. The connections view draws the asset as a graph, drug product to container closure to quality control, filterable by domain. The data room keeps the regulatory sources in a plain file tree with the PDF open right beside it, so checking a citation never means leaving the platform.

Data connections view: a React Flow graph of asset entities including drug product, container closure systems, and quality control nodes, with domain filter chips along the top — Fig. 03Data connections. The asset's entities as a filterable graph, built on React Flow.

Virtual data room: a regulatory folder tree on the left and an inline PDF viewer on the right showing an MHRA questions document — Fig. 04The virtual data room. Regulatory file tree beside an inline viewer, here an MHRA questions document.

Documents that draft themselves

Regulatory and commercial documents are where drug programs lose their weeks. The document builder starts from the asset's own data, drafts the whole thing, and then walks the author through three honest stages: generate, iterate, finishing touches. The report shown here runs 32 pages. Starting one takes a description and whatever reference files you want it to lean on.

AI document builder showing a 32-page Market Scan Report open in an editor, with a stepper tracking the generate, iterate, and finishing-touches stages — Fig. 05The document builder mid-flight on a 32-page market scan report.

Create New Document modal with a free-text description field and an upload area for reference files — Fig. 06New documents start small: a description and a stack of reference files.

Trust built in

Drug assets change hands, and a buyer inherits the record along with the science. Every consequential action on the platform is committed to a ledger, and the explorer presents those blocks as plain accounting rather than cryptography theater. Conversation stays inside the same walls too: asset-scoped channels and direct messages, carried by Matrix, living next to the data they discuss.

Blockchain explorer with transaction statistics, a breakdown of operations by type, and a list of ledger blocks with their hashes — Fig. 07The blockchain explorer. Operations broken down by type, ledger blocks and hashes below.

Messages view with asset-scoped channels and direct messages in a sidebar and an open conversation thread — Fig. 08Messages. Channels are scoped to the asset, so context never needs re-explaining.

The whole portfolio at a glance

Leadership asks a different question: not how one program is doing, but what the whole pipeline is worth. Scenario modeling lays every asset on a single timeline that runs out to 2067, with net present value attached to each program. Drag a phase and the downstream numbers follow. It turns a spreadsheet argument into something you can point at.

Portfolio scenario modeling: a multi-asset gantt timeline reaching to 2067, with NPV figures per asset and a tooltip open on one development phase — Fig. 09Portfolio scenario modeling. Every asset on one timeline to 2067, NPV attached, phases draggable.

System notes

Bone paper canvasOne accent · oxidized orangeTables tuned for scanningDark mode variant shipped