Scientific Expression Builder
An expression builder for pharmaceutical batch records to ensure accurate calculations.
- Company
- MasterControl
- Product
- MX Production Records & Logbooks
- Role
- UX/UI Designer
- Team
- PM · Researcher · Tech lead · PO · Designer
- Timeline
- May 2024 — Q3 2026
Context
Calculations are 10–80% of any electronic batch record. Most are still done in spreadsheets and transcribed back — two chances to fail per value.
By the numbers
- 90%
- Target reduction in calculation deviations
- 25%
- Target reduction in batch-record review time
- 3–5K
- Template nodes the builder must validate
Problem
01 / 05
The engine was powerful; the interface wasn't. Authors couldn't see what variables existed, couldn't predict what would work, and couldn't test before shipping. So they used Excel and pasted the result.
"How might authors build, test, and debug formulas with confidence — without falling back to spreadsheets?"
- →Discoverability — what variables and functions exist
- →Debuggability — errors that point at a fix
- →Testing — validate before production
- →Consistency — same variable, same behavior, everywhere
Research
02 / 05
Customer expression data, PM and CS interviews, journey mapping across four personas, and external benchmarks (EY, Rockwell, FDA). 8–10 participants pressure-test the design in Q1 2026.
Principles
03 / 05
Five rules for hard trade-offs.
- →Clarity over cleverness
- →Prevention through testing
- →Progressive power — GUI for simple, type-ahead for fluent
- →Actionable feedback on every error
- →Consistency is trust
Solution
04 / 05
Two authoring modes on one validated model — type-ahead for power users, GUI for visual ones — plus simulation that proves a formula works before it ships.
- →Type-ahead editor with inline variable type and step context
- →GUI builder: grouped variables, categorized operators, live preview
- →Simulation with auto-generated inputs and edge-case tests
- →Categorized errors with offending token underlined; auto-fix where safe
- →N/A handling and circular-dependency prevention across 3–5K nodes
Status
05 / 05
Design complete; engineering aligned. Phase 1 beta Q1–Q2 2026; Phase 2 simulation in Q2; Phase 3 templates and conditional logic in Q3.
Impact
- 01
Reframed a low-adoption tool around two modes sharing one model
- 02
Made testing a first-class authoring step — simulation didn't exist before
- 03
Targeting 25% fewer batch deviations via a 90% drop in calc errors
Reflection
"The hardest part wasn't the editor — it was being honest about what we'd only know once users got their hands on it."