A proprietary AI toolchain. A human at the wheel.
Math modelling at Hip Durian runs on an entirely proprietary AI toolchain — a bespoke model architecture we designed for slot mathematics, hosted on a dedicated network of GPUs. Every math model is written by a human, reviewed by a human, and pressure-tested through billions of simulated spins before art is committed.
This is the part of the workflow we don't publish and don't outsource. It's the difference between studios that wrap a public model in a wrapper and studios that own the primitive.
The tool doesn't design the game. It lets a single mathematician iterate at the speed of ten. Feli — our founder, engineer, and mathematician — can propose a math variant in the morning, simulate a billion spins over lunch, and have a signed-off release candidate by end of day.
Tentative final math is up. Base game is 9.5/10 in terms of volatility (coefficient of variability). It's defined as VERY HIGH. Feature game is 8.5/10. Ran 9 hours of simulation on 100M spins. One release candidate to Edgar (our external math reviewer) as a check.
// Hold Yer Gold — release candidate 1
{
"game": "Hold Yer Gold",
"candidate": "RC1",
"sim_spins": 100000000,
"sim_hours": 9,
"rtp": 94.7,
"volatility": 9.5,
"hit_freq": 0.243,
"max_win": 5000.0,
"variant": "sweeps"
}The proprietary edge: variants at speed
The tool's real power isn't the single-model run. It's what happens when you need two math models in parallel — one for a WAYS-style pay evaluation, one for a LINES-style pay evaluation, both tuned to the same target RTP and volatility band.
On 3 Purramid Nights, we shipped both variants to David for internal preference testing before art was locked. Not "we could ship both eventually" — we shipped both, in the same week, from the same math bench.
Current vs proposed on 3PN modifier coins: BOOST +10 gives own orb 10× base bet. MULTI ×4 gives 4× base bet. Initial BOOST capped at 10× but accumulated prize can grow. Overall 10,000× win cap. Two variants running side by side — WAYS and LINES. Preference?
What the tool doesn't do
It doesn't invent the game. It doesn't decide the theme. It doesn't pick the volatility band. Those are human calls — informed by operator conversations, market gaps, and player-behaviour data.
What it does do is make the maths step, historically the bottleneck of every slot studio, no longer the bottleneck.
Proprietary
Bespoke architecture we designed and own. Not a fine-tuned public model.
Hosted on our GPUs
A dedicated network. Not a shared cloud endpoint someone else can throttle.
Human-authored
Every math model is written and reviewed by a human before it ships.