Our tools take a model past the known: memory of your work from one session to the next, knowledge drawn from your own documents, and a safe home to work in.
Andrew Grainger builds AI systems with a scientist’s discipline
His studio, Voidwright, makes tools to fit into projects you already run.
Research gave Andrew a way of working he brings to any field
He enters unknown ground with confidence, questions what he finds, and plans each step with an experimenter’s rigor.
He tested his deep-learning method against expert annotation
Andrew trained as a scientist, with a PhD in biochemistry and molecular genetics. His AI work began in research. As a postdoctoral researcher, he developed and published a deep-learning method that measures fat on CT scans, evaluated against expert annotation.
Confidence carried him into one field after another
His move from research took him beyond his doctoral training, into the commercial life-science industry. There he supported scientific instruments where they were in use, and then led product marketing for instrument lines.
The marketing engine he designed puts market research before the message
He designed and built an AI-native, full-stack marketing engine. Market and competitor intelligence feeds the positioning and messaging, which feeds content in many forms, such as blogs, podcasts, technical guides, landing pages, social posts, paid ads, and videos. Automated distribution carries that content across channels, including a website and resource hub he also built. Technical and scientific checks run before anything goes out.
His retrieval layer cites sources; code, not models, writes his intake’s records
He built a document retrieval layer whose results carry their sources, connectors that let AI agents query scholarly literature and other services, and AI-assisted intake into a company’s live records, where code, not the model, owns every change. He also trained and evaluated models. Andrew co-founded a small agency, where he leads AI engineering; Voidwright is separate from that agency.
He maps a system before he connects it
Before connecting systems, Andrew maps who acts, what data moves, which permissions apply, and where anything is written. He tests in stages and keeps every source traceable. A person’s review comes before release. His work on AI systems is directed and reviewed engineering: he designs the systems, directs and reviews the code, and hardens it.
Voidwright builds tools for the model you choose
Andrew weighed his studio’s course as he weighs any problem, and chose to augment others’ models.
We do not make AI models. We build the tools that make the ones you already use more useful. Each tool is aimed at what a model lacks on its own. Voidloom gives it memory of your work across sessions. Voidmill gives it knowledge of your documents. Voidstead, which is being built, would give it a safe home on your machine. None of this has been benchmarked.
We chose to build the parts that keep their value when a model is replaced, because the leading model changes month to month, and also because frontier labs invest at a scale no studio matches. Voidstead is a home designed around that choice, with everything of value in vendor-neutral files you hold. It would treat the model, hosted or local, as a replaceable part, measured rather than trusted. That home has not yet been tried with a live provider.
Working together could run in either direction
Beyond the studio’s tools, Andrew does hands-on work building AI systems into existing projects, which could include yours. If you are building a product, he could lend Voidstead, a local workspace for AI agents, as that product’s harness. Without being rewritten, Voidloom or Voidmill could also become one part of a project already running. If you have built skills and automations for one kind of work, your own pack could bring them into Voidstead; packs are planned.
Questions about Andrew Grainger and Voidwright
Who is behind Voidwright?
Andrew Grainger is behind Voidwright: it is his studio, and today nobody else works in it. A scientist by training, he designs and builds AI systems.
What is Andrew Grainger’s background?
Andrew Grainger holds a PhD in biochemistry and molecular genetics. In postdoctoral research, he published a deep-learning method for CT scans. He moved into life-science instruments and product marketing. His AI systems span marketing, document retrieval, and record intake. He now leads AI engineering at a small agency he co-founded.
Why does Voidwright build tools around AI models rather than build models?
Because Voidwright decided to build what still holds value after one model replaces another. Which model is ahead shifts from month to month. Companies that make frontier models spend at a level no studio can match. The tools have not been benchmarked.
How could someone work with Voidwright?
Andrew Grainger’s own engineering is on offer: someone could ask him to build AI systems into an ongoing project. Voidloom and Voidmill are not yet offered to others; if offered, either could serve as one piece of someone’s existing work. Voidstead, a workspace designed for AI agents, is being built. Someone could later borrow it as their product’s harness, or gather skills and automations into their own pack for Voidstead, where packs are planned.
Voidwright is Andrew Grainger’s studio, and today he is the only person behind it. We build Voidloom for agent memory and Voidmill for document knowledge. We are building Voidstead to bring those tools into one home. Working with us could mean integrating a product into an existing project, as a piece of something else, or bringing a collaborator’s work into that home. Neither would mean rewriting the products. Integration work can span fields, but the models Voidmill draws on currently serve life-science and scientific literature. Applying that document stack to other subjects would require retuning, which we have not promised.
Your project could take in one of these tools, or your work could come into Voidstead.
Voidloom and Voidmill are in private use; Voidstead is in development; none is available to others yet.
As of , Voidloom’s background thinking runs on the hosted or local model chosen at Voidloom’s own setup, if any. With a local model, research into gaps in what Voidloom knows is skipped because that research needs the web.