Skip to content
Xplore
Reliability infrastructure for AI agents

Agents enterprises
can trust in production.

The evaluation, training and control layer that makes AI agents safe for regulated, high-stakes work — where mistakes are not allowed.

Forge
Private full-lifecycle training & control
Agent 007
Public real-world agent benchmarks
The problem

The black box.

The costliest errors happen in regulated, high-stakes work — exactly where agents still can't go. They hallucinate, ignore instructions, and the failure is untraceable. If you can't see it, you can't fix it, and you can't trust it.

$47B
Agentic-AI market by 2030 — concentrated in regulated work
70+
Enterprise interviews naming the same blocker: the black box
1
Invented step, hidden among thousands, breaks the whole result
The solution

Train the errors out — measured, not promised.

A data protocol makes every reasoning step and action visible — turning an opaque agent into something measurable, traceable, and trainable.

01
Protocol
Every step made visible
02
Train
Find, fix, re-measure
03
Control
Reliable after model change
How the platform works →
Proof

A live agent, shipped in weeks.

Wellbeing-risk ECG agent, built on Forge and shipped inside a client's product.

40+errors caught pre-prod
£2M+losses avoided
30–100%error reduction
£2.5MTCV over 5 years
Traction · last 90 days

A funnel converting into revenue.

4 in active sales
2 in contract scoping
1 signed · £144k ARR
Founded UK patent filed SFC grant Domain pilots Public cases live Breakout 50
The ask

£1.5M — SEIS/EIS, two tranches.

£400k
Now
£1.1M
Next
Convert the funnel — pilots into contracts.
Scale platform, R&D, and main-domain expansion.
Tranche 2 unlocks at 3–4 signed contracts (~£300k ARR) — de-risked by revenue, not promises.
The moat

Defended on four fronts.

The market is converging on the same primitives — evals, traces, observability. We own the layer that closes them into one loop.

01 Data protocol patent filed

Makes every reasoning step and action visible — the thing that turns a black box into a trainable system. Nobody else has it.

02 SimEngine · IS/OS/LIVE

One design spans training data, held-out verification, and live monitoring — generating tasks, ground truth and traps from client data.

03 Case-specific trainers

Strategies mapped to task families. Model- and runtime-agnostic: the model stays replaceable, the trained configuration is the asset.

04 Data flywheel

Every public case and private contract grows a proprietary corpus of agent failure modes. Benchmarks drive signal; contracts compound data.

Licence recurring platform fee — sticky once embedded in a regulated workflow
Usage per-run training + per-query inference — scales with adoption
Team
Alex Shkrebelo
CEO
GTM & partnerships · 2 early exits · UK Global Talent
George Izgarshev
CTO
AI, data engineering, graphs · 2 exits · Forbes 30u30
Sikko Zoer
Advisor
ex-VP Global Logistics, Medtronic (17y) · ex-BCG
Request the data room