The company

Substrate

AI agents for the physical world.

BuildingProblem validatedPre-seed
The problem

Every great AI company is being built for software developers. The engineers who design the physical world were left behind.

Chips, circuits, embedded systems, hardware — the foundations everything else runs on.

The people who design them still work in tools from the 1990s. They read schematics by hand, trace nets across pages from memory, and modernise decades-old designs with almost no leverage from the AI revolution happening one floor up.

It isn't a small niche. It's the entire substrate beneath modern technology — and it's been structurally skipped.

Why me

Building this needs someone fluent in both worlds. Almost no one is.

To build AI for hardware you have to actually understand electronics and modern machine learning. Those two populations barely overlap.

I'm a second-year Electronic Engineering student who has spent two years building and shipping AI products. That combination is the whole thesis. The software-AI world is building for software; the hardware world can't build the tools it wishes existed. I sit on both sides of that wall.

Read: The knowledge wall
The wedge

Tracer — a verifiable copilot for legacy circuit modernisation.

A wedge you can win beats a platform you can only describe. Tracer helps engineers modernise legacy electronic circuits step by step — with the engineer checking every output.

01

Read the circuit

Tracer ingests a legacy schematic and infers what each component is, what the sub-circuits do, and what the whole design is functionally for. This understanding layer is the genuinely hard part — and the primitive my team built at the AMD contest.

02

Propose, step by step

It suggests a modernisation one verifiable step at a time — never an autonomous black box that multiplies errors in high-stakes work.

03

Engineer verifies

Each output is inspectable and reversible. The engineer stays in command; trust compounds with every confirmed step.

04

Ships inside their tools

Tracer lands as an assistant inside the EDA tools engineers already use — Altium, KiCad, Cadence. Low friction, no behaviour change required.

Why now

Three things are true at once.

01

Models can finally see

Frontier multimodal models can now read schematics end to end. The inputs are no longer the bottleneck they were even a year ago.

02

Budgets are urgent

Defence reshoring and supply-chain mandates have created real, time-pressured budgets for modernising legacy electronics.

03

Engineers are ready

Every engineer I interviewed wants a tool that works alongside them — not one that replaces their judgement. The demand is shaped and waiting.

The path

From wedge to platform.

Now

Tracer

A verifiable modernisation copilot inside engineers' existing EDA tools. Win the wedge; earn the trust and the data.

Next

Modernisation EDA

Tracer grows into a purpose-built modernisation platform — the foundation other Substrate agents are built on.

Then

The Substrate agents

Schematic, Embedded, and beyond — a family of agents for everyone who designs the physical world.

Schematic · Embedded · …Future agents
Modernisation EDA platformThe foundation
Tracer — verifiable copilotThe wedge · today
Circuit-understanding primitiveBuilt at AMD · 2nd of 100+
The compounding moat

Every engagement makes the next one harder to beat.

Each modernisation Tracer touches adds proprietary design data and engineer trust that the next layer inherits.

Data the rest of the market can't get. Trust a black box can't earn. The platform doesn't just grow — it gets structurally more defensible with use.

Where it stands
Stage
Pre-seed — building Tracer MVP
Proof
Core primitive built at the AMD Pervasive AI Contest — 2nd of 100+ teams
Validation
Engineer interviews across modernisation, defence & biomedical
Status
Problem validated. Building the MVP and talking to founders
The bet

What Cursor is to software, Substrate is being built to be for everyone who designs the physical world.