Probity Point Tools to enable people Sheet 01 of 04

I build tools that help people do things.

For twenty years I have developed validation and visualization software for structural biologists. I am branching out into making tools for other fields. Every tool I have made is designed to help people more easily accomplish something. Where are the mistakes in this structure? Does this food have ingredients I cannot eat? How do I use technology to improve my hockey skills?

I will not promise software without bugs. Nobody honestly can, least of all now that a machine writes more code in an afternoon than a person can read. What I promise is a tool you can trust when it tells you something you did not want to hear.

Principal Vincent Chen Scientific software since 2004 Incorporated 2026 Chapel Hill, NC vincent@probitypoint.org
01

Work

JOB 001

MolProbity Next

Structure validation, rewritten

A complete rewrite of the MolProbity validation service, using modern web technology. The validation runs behind a task queue, so a structure that takes minutes to validate does not hold the browser. Results land in a sortable per-residue table beside three viewers: Mol* to look, Moorhen to correct, and Kinemage for probe dots and validation markup, with interactivity between the results table and the viewers.

Each residue carries one MP-Triage score answering which to fix next, driven by its worst finding rather than the sum, because four small clashes should not outrank a twisted peptide. It is reported to one decimal place so nobody reads it as a count.

Field note One part I find especially useful was not originally planned. All three viewers sit on the same result, so you can switch representations without losing your place, and different views make different problems obvious. It came about during development; I noticed its power and kept it.
Stack
Django, CCTBX, PostgreSQL, Mol*, Moorhen, Kinemage
Scale
56,385 lines
1,512 unit tests
26 E2E specs
346 commits
Period
2025.08 to present
Status
In active development
Fig. 1 — The compact view, ordered by MP-Triage. Clicking a row drives the viewer beside it. Entry identifier withheld.
A residue table ordered by triage score, the worst rows tagged with
                    the issues found and the hotspot they belong to, beside the kinemage
                    viewer zoomed onto a clash and drawing the contact spikes.
JOB 002

AisleAlly

A label reader that refuses to guess

I know a lot of people with food restrictions, and the big allergen classes in the US do not cover most of what they avoid. Database tags only track the big nine, so anything else has to come from the ingredient text. Other apps I looked at had subscription fees, which I wanted to avoid.

AisleAlly reads that text against more than a dozen classes outside the big nine, from sulfites and nitrites to MSG and nightshades. Green is clear, red a direct hit, yellow when an umbrella term like "spices" or 香辛料 could be hiding one. Lookups reach a vision model only as a last resort, which is what keeps it subscription-free.

Fig. 2 — Verdicts across the catalogue, one avoidance class
Clear 1,606 Cannot tell 8,602 Contains 3,980
Field note More than half the catalogue comes back yellow, which is the design working. A two-state verdict would have to manufacture confidence the ingredient text does not support, and a confident wrong answer is the one thing somebody avoiding an ingredient cannot use.
Stack
React, Vite, Cloudflare Pages and Workers, vision model OCR
Catalogue
15,389 grocery
1,122 restaurant
Tested
124 cases
Status
Live at aisleally.pages.dev
JOB 003

HockeyHandler

A stickhandling trainer that watches the puck

Built by a beer-leaguer, for beer-leaguers and up. The trainers I looked at either want a lot up front or bill through the off-season. I wanted something instructive at a price that makes sense for a game most of us play once a week.

The best part is the puck tracker. Set up a phone camera over a shooting board and it guides you through more than a dozen stickhandling drills while it tracks the puck, so off-ice practice stops being something you wonder if you are doing correctly.

A second device displays the drill, keeping it at eye level. Stickhandling with your head down is the habit you are trying to break.

Around it sits the rest: reaction and timing games for skaters and goalies, a drill library you can build into timed routines, a coaching diagram editor, and a 3 v 3 arcade game.

Field note Every session records results, so the app can tell you whether you are getting faster across months rather than just how you did tonight. A trainer that reports only tonight is a stopwatch, and the trend only exists if every drill recorded the same way from the first one.
Stack
Swift, SwiftUI, SwiftData, AVFoundation
Tested
Drill scoring and board calibration, under unit test
Access
Screen reader, largest text sizes, reduced motion
Status
Preparing App Store submission
02

Method

Billing

You always know where the meter is

A fixed price is an estimate with the uncertainty hidden inside it, so I bill for time instead, keep the hours visible as they accrue, and stay flexible about the arrangement.

Staffing

The people who scope it are the people who build it

There are two of us, both out of the same lab, with no bench, no handoff, and no account manager between you and the code.

Tooling

A toolsmith succeeds only as his users do

Fred Brooks said that here in Chapel Hill, and it applies twice now: my tools amplify the people using them, and AI amplifies me, while my twenty years of scientific software experience tells a good AI answer from a merely plausible or an erroneous one.

Handover

Your repository, your infrastructure, your documentation

Tests and written decisions come with the code, so the work can be read, checked, and continued without me.

03

Principal

Vincent Chen
Principal Consultant

I joined the Richardson lab at Duke in 2004 and stayed for a PhD, spent several years at UW Madison, and returned to Duke in 2016. The work has mostly been tools that other scientists depend on to do their own: MolProbity for structure validation, KiNG for visualization, and Phenix for structure determination.

Scientific software gets used by people who notice when an answer is quietly wrong and publish what they found, which is good training for distrusting anything that merely looks correct.

I work with one other engineer at roughly half time, also out of the Richardson lab, whose strengths run where mine do not. Between us there is nobody to hand your project down to.

Selected publications · every entry links to PubMed
2025
Categorizing prediction modes within low-pLDDT regions of AlphaFold2 structures: near-predictive, pseudostructure and barbed wire
Williams CJ, Chen VB, Richardson DC, Richardson JS · Acta Crystallogr D Struct Biol 81(10):558-572 · PMID 40937679
2021
Improving SARS-CoV-2 structures: peer review by early coordinate release
Croll TI, Williams CJ, Chen VB, Richardson DC, Richardson JS · Biophys J 120(6):1085-1096 · PMID 33460600
2015
MolProbity for the masses-of data
Chen VB, Wedell JR, Wenger RK, Ulrich EL, Markley JL · J Biomol NMR 63(1):77-83 · PMID 26195077
2010
PHENIX: a comprehensive Python-based system for macromolecular structure solution
Adams PD, Afonine PV, Bunkóczi G, Chen VB, and 14 others · Acta Crystallogr D 66(2):213-221 · PMID 20124702
2010
MolProbity: all-atom structure validation for macromolecular crystallography
Chen VB, Arendall WB 3rd, Headd JJ, Keedy DA, Immormino RM, Kapral GJ, Murray LW, Richardson JS, Richardson DC · Acta Crystallogr D 66(1):12-21 · PMID 20057044
2009
KiNG (Kinemage, Next Generation): a versatile interactive molecular and scientific visualization program
Chen VB, Davis IW, Richardson DC · Protein Sci 18(11):2403-2409 · PMID 19768809
04

Contact

vincent@probitypoint.org Replies within one business day