Pilot program · 2027

Twelve weeks on one of your targets

We are looking for three design partners: R&D teams with a clear materials target, some reference data, and a lab that can make a dozen samples. We bring the platform, the GPUs and two scientists.

Abstract

Partners
3
Length
12 weeks
Start
2027
Terms
Draft, to be agreed

01 · Benefits

What design partners get

Science

Two Forgite scientists

Working alongside your computational chemists for the whole pilot.

Compute

GPU budget included

Enough GPU-hours for one full round on your space.

Models

A fine-tune on your data

An Equiva potential fine-tuned for your chemistry, kept in your tenant.

Influence

Shape the product

Your workflow sets the order of integrations and features.

02 · Plan

Week by week

  1. 1

    Weeks 1–2 · Scope

    Agree the target, search space, reference data and the held-out test.

    Output Written plan
  2. 2

    Weeks 3–4 · Fine-tune

    Fine-tune and validate the potential on your data.

    Output Validation report
  3. 3

    Weeks 5–7 · Screen

    Run the screen and MD; review the ranking with your team.

    Output Ranked list
  4. 4

    Weeks 8–11 · Make

    Your lab or a partner makes and measures up to 12 candidates.

    Output Measurements
  5. 5

    Week 12 · Review

    Fine-tune on the results and write up what worked and what did not.

    Output Pilot report

03 · Fit

Who the pilot suits

Good fit

  • A target property with units and a limit
  • Some DFT or experimental reference data
  • A lab that can make about 12 samples in a month
  • A named person who can approve batches

Not yet a fit

  • Polymers and amorphous materials
  • Targets that need excited-state physics
  • Programs with no lab access at all

04 · Program form

Put your program forward

Rough numbers are fine.

05 · Questions

Pilot questions

What does the pilot cost?

Terms are a draft. We expect a reduced fee that covers compute, in exchange for feedback and a case review that you approve before anything is published.

Who owns the results?

You do: your data, your fine-tuned model and any materials you discover. We ask only to describe the method, with your approval.

What if the screen finds nothing useful?

Then the report says so, with the reasons. A clear negative result in twelve weeks is still cheaper than a year of trial and error.