MCS Strategies LLC

Selected work

Thirty years of software that created markets.

From the first desktop 3D modeling software for drug design to AI-native builds in 2026. For current products, see Answer Reel and PillSheet.

Built with AI, 2025–2026

Expert-checked, not just plausible.

I choose problems where I can verify every answer. That is how I learned what AI-built software can be trusted to do.

Sculpt rebuilt: a protein rendered with ribbons, atoms, and hydrogen bonds on a dark background.

Sculpt, rebuilt

In 1992, for my Ph.D. at UNC–Chapel Hill, I built Sculpt: a chemist grabs atoms on screen and the protein relaxes around the change in real time, under a full molecular-mechanics force field. Sculpt made the cover of Science and became Interactive Simulations, the company I founded and sold to Reed Elsevier.

In May 2026 I rebuilt it from the two original papers with AI coding agents, in about a week. Few people can check thirty-year-old scientific software line by line. Its author can, which made it the ideal benchmark.

Inside
CHARMM19 force field, sparse constraint projection, Richardson ribbons, hydrogen bonds, contact dots, real-time GPU rendering.
Source
Surles, Richardson, Richardson & Brooks, Protein Science 3(2), 1994.

Long-document Q&A

Clinical teams live inside 100+ page trial protocols, where the answer to one question spans a multi-page table, three sections, and an amendment. Generic chat tools lose that structure. This system keeps it: an AI agent navigates the document's own section tree and answers with clickable, page-anchored citations.

I built it twice: by hand in 2025 (custom retrieval, structure-preserving parsing, specialist agents per question type), then rebuilt API-first in 2026 with an evaluation set and a latency budget.

Built as
An embeddable API for document-heavy products, plus a review UI.

ScienceMedia, 2011–2024

Two products that rebuilt a company.

As CEO I also led product development. With the ScienceMedia team I created these two cloud products, which turned a services firm into a product company with five-fold revenue growth.

SMi Source on a phone: disease-education lessons by therapeutic area.

SMi Source

The largest disease-education library for life science professionals: short expert video topics on any device, about 1,000,000 views a year at 92% gross margin.

ScienceMediaSaaSVideo learning
SMi Trial on a tablet: clinical trial protocol training.

SMi Trial

Created the market for risk-based clinical protocol training: fewer costly operational mistakes, better patient safety, and faster regulatory approval.

ScienceMediaClinical trialsCompliance

SMi Source and SMi Trial are products of ScienceMedia, Inc.; images courtesy of ScienceMedia.

Also built

Video, documents, data, and decisions.

Patient video pipeline

Multi-agent pipeline that turns clinician training video into patient-friendly education: script, visuals, voice-over, and assembly. The forerunner of Answer Reel.

KitchenGPT

An OpenAI custom GPT backed by an evolving web database: a household cookbook that remembers what was cooked and what worked. Built with a different AI coding agent, to compare toolchains head to head.

Financial modeling

Monte Carlo forecasting, tax optimization, and scenario engines, checked line by line against professional results. The same rigor I bring to pricing, market sizing, and cash-flow models.

How I build

Fast, measured, and ready for diligence.

  • Service first. One business-logic layer behind an API; every UI is a client of it.
  • Measured. Latency budgets and an evaluation harness from day one, so every change is graded, not guessed.
  • Model-agnostic. OpenAI, Anthropic, and Google models A/B-tested behind one interface, with local options where data must stay in-house.
  • Audit-ready. System cards, data-flow tables, and acceptance criteria a buyer's diligence team expects.

Start a conversation