Computation-al capability buildout

Context

In an early-stage biotech environment, I joined as the first computational hire to help establish the infrastructure, workflows, and analytical foundations for a new discovery programme. The challenge was to build a credible operating capability quickly while working under significant time and business pressure.

The Problem

The core tension was speed versus rigour. Building a computational drug discovery platform properly takes time — defining the right architecture, validating analytical choices, ensuring results are reproducible and defensible. But the business needed a working proof-of-concept and a compelling investor story within months. Underneath both sat a practical bottleneck: even if the ML platform generated promising drug target predictions, biologists needed to validate them experimentally — a process that at the time took three weeks per screen to analyse manually.

What I Did

The first task was infrastructure — defining requirements for compute, tooling, and analytical workflows from scratch, and architecting the cloud environment that would underpin all downstream work. In parallel, I worked with the founders and biology team to define a focused product angle grounded in where the company had genuine scientific and operational advantages, translating that into a product vision and roadmap. The third thread was the validation bottleneck. Automating the screening analysis pipeline brought turnaround down from three weeks to 30 minutes, turning a slow sequential handoff into a faster feedback loop for the biology team.

The Outcome

Automating the screening pipeline brought validation turnaround down from three weeks to 30 minutes, enabling a much tighter iteration cycle between ML predictions and experimental validation. Together, the platform supported proof-of-concept work across multiple disease indications and contributed to a meaningful funding milestone within 18 months.

What This Illustrates

Building a platform from zero in a fast-moving environment means making early decisions that are hard to reverse — about architecture, priorities, and where to invest limited time. The validation bottleneck was a good example of a problem that wasn't immediately obvious but turned out to be central: without fixing it, the iteration cycle between prediction and experimental confirmation would have been too slow to be useful.