Epinodal — Investors

Why now

The incumbents just admitted the gap. We started two years ago.

Models finally have the headroom for assistant-grade intelligence — but on their own they still can't learn. They document; they don't consolidate; and eventually the context gets too large to reason over. We built the system that closes that gap: an AI that identifies new information and teaches itself.

The breakthrough

Learns new knowledge without catastrophic forgetting.

A preservation-first learning architecture — neuroplasticity-inspired research IP, running on a production enterprise agent platform. The research makes it learn; the platform makes it deployable.

92.3%

vs 55.1% — conventional LoRA

Knowledge retained

After learning a new task in a controlled continual-learning study, where the conventional approach degraded ~45%.

10/10

vs 0/10 — strongest control

Ordinary behaviour preserved

At LLM scale, Epinodal kept ordinary prompts correct where the fine-tuned control had broken them.

0.20

vs 0.70–0.95 — controls

Domain contamination

A clean recall-vs-preservation balance — learns the new domain without bleeding into the rest.

Figures from controlled, in-repo studies (a tiny-MLP continual-learning study and an LLM-scale Gemma run), not a universal benchmark. Part of this raise scales the infrastructure to stress-test the upper limits — we haven't hit the ceiling yet.

Traction

Real third-party validation, not a demo.

A paid enterprise pilot is live — continuous-learning AI project management, with expansion gated on outcome.
A second opportunity is in motion in healthcare — a digital front-door agent for an enterprise IT project office.
Primary route to market: AWS Marketplace — selling internationally through the founders' existing AWS relationship.
University research relationship in place on the continuous-learning models.

The ask

Raising $1.5M AUD (seed).

To prove the learning ceiling, secure the IP, and turn a global distribution channel into the sales engine.

$250K

Infrastructure — scale the horizontally-scaling stack and stress-test the learning ceiling.

$50K

Patents & IP — file on the continual-learning architecture.

$200K

Global go-to-market — AWS Marketplace roll-out.

~$1.0M

Freemium-led client acquisition and international company formation.

Why this team

Research that knows how memory works, married to a platform already through enterprise review.

James Antoniadis

Co-founder · Learning Research

Practising psychiatrist and neural researcher. Years studying how human memory actually works — the brain-plasticity theory that makes genuine continuous learning possible, not probabilistic guessing. The learning model is his domain.

Bruce Lock

Co-founder · Platform & Enterprise Delivery

Built a production AI agent platform — agent harness, dynamic decision engine, governed cheapest-capable model routing. Won an enterprise innovation program against 700 teams, and got agentic platforms live through responsible-AI and security review in Fortune 500 environments.

We re-platformed for AI that learns. Now we're raising to prove its limits.