project25.live
SYSTEM HEALTHNOMINAL

The control plane for attention.

Project25 generates content, publishes it from a fleet of 1,968 real devices, measures exactly what earned attention, and rewrites the next batch. A closed loop that gets smarter every cycle, with no human in the path.

145.9M+18.4%
30.9M+9.1%
~4.4M-2.3%
1,350+6.0%
153+3.2%
104+27.5%
VIEWS / MIN3,041upPOSTS LIVE1,350upACCOUNTS ONLINE153FLEET UPTIME99.9%upAVG PUBLISH1.4sdownDEVICES1,968METROS42QUEUE DEPTH86downERROR RATE0.03%downGPU NODES6 / 8RENDERS / 24H12,407upPUBLISH SUCCESS99.97%up
SYSTEM ARCHITECTURE

The closed loop system.

No human touches a post between idea and impression. Generate → Review → Publish → Measure → Learn. One closed loop, fully instrumented.

CLOSED LOOPno human in the path
GENERATEThousands of variants per cycle~3K / CYCLE
REVIEWQC, policy-risk + performance score0 HUMANS
PUBLISHWinners ship to the live fleet1,968 DEVICES
MEASUREEvery impression attributed backPER-IMPRESSION
LEARNResults rewrite the modelAUTO-REWRITE
PERSONA LIFECYCLE

Personas that perform, and recover when they drift.

Each persona has a name, niche, voice, and device. They move through states, and auto-recover when performance drifts.

Luna ByteTech Tips · #0147
warming
voiceConfident · Calm
deviceiPhone 14 Pro
cadence2 / day
+12%30D VIEWS
Aria ByteTech Explainers · #0231
active
voicePrecise · Fast
deviceVMOS · pad-08
cadence3 / day
+184%30D VIEWS
Fit AtlasFitness · #0388
recovered
voiceDirect · Warm
deviceGeeLark · g-41
cadence2 / day
+78%30D VIEWS
createwarm uppublishmeasurerecoverpersonas auto-recover when performance drifts
PERFORMANCE & OBSERVABILITY

Performance that compounds.

Every post teaches the next one. Real-time pipeline, full visibility — every signal fed back into the model.

248
QUEUE
221
QC
207
CLEAN
186
LIVE
83.5%PASS RATE
PERFORMANCE OVERVIEWVIEWS / DAY · 30D
30D VIEWS145.9M+18.4%
PEAK DAY10.0M+27.5%
AVG / DAY~4.4M-2.3%
TOP ACCOUNTS30D VIEWS
01Luna Vegalifestyle1.29M
02Mara Quinnfitness1.10M
03Aria Stonetech969K
04Cleo Marshbeauty813K
05Nova Reyestravel744K
06Iris Lanefood690K
07Sage Bennetwellness613K
08Remy Colefinance558K
09Juno Pikegaming492K
10Vera Solfashion432K
VIEWS / DAYLAST 14D
FLEET HEALTHUPTIME
99.9%NOMINAL
  • EXCELLENT128
  • GOOD19
  • DEGRADED4
  • OFFLINE2
COMPUTE CLUSTERGPU NODES
153DEVICES
6 / 8GPU NODES HOT
~104VIDEOS / MONTH
99.9%CLUSTER UPTIME
LIVE SYSTEM LOGproject25 · fleet · tail -f
02:14:03 dispatch · job q8f3a → device pad-08 · queued
02:14:05 render · seedance2_fast 720p · 4.1s · ok
02:14:07 publish · clip rk_92f1 → channel-01 · 200 ok
02:14:09 measure · +12,407 views · 24h window
DISTRIBUTION

Publish everywhere attention lives.

A live fleet of 1,968 real devices, kept alive against adversarial platforms at 99.9% uptime, across every metro that matters.

1,968DEVICES ONLINE
99.97%PUBLISH SUCCESS
42METROS
1.4sAVG PUBLISH
VERTICAL VIDEOShort-form · algorithmic feed812 devices online
DISCOVERY FEEDRecommended surface1,156 devices online
PROOF

Numbers that hold up.

PRICING

Simple pricing. Built for scale.

You pay for video, not seats.

STARTER$0

1 profile, 50 renders

  • 1 synthetic persona
  • 50 video renders / month
  • Basic analytics
  • Community support
GET STARTED
SCALEcustom

dedicated GPU, unlimited profiles, SLA

  • Unlimited personas
  • Dedicated GPU cluster
  • Custom throughput
  • SLA & uptime guarantees
  • Dedicated support
CONTACT SALES
BUILT BY ENGINEERS

Built by a small technical team.

Mathematics and computer science (Oxford, UCL, HSE), with a track record of AI apps reaching millions of users. Based in Oxford, since 2025.

OXFORD
UCL
HSE
EST. 2025 · OXFORD, UK
FOUNDERS4
DISCIPLINEMathematics + Computer Science
UNIVERSITIESOxford · UCL · HSE
TRACK RECORDAI apps → millions of users
BUILD WITH US

The hard part is the product.

We keep the team small on purpose: every engineer owns a layer of the loop, end to end. No managers, no busywork. Real-world adversarial systems work, shipped daily. Four people, 1.2B views.

OPEN PROBLEMS
Survival

Keep ~1,968 real devices alive against adversarial platform detection: fingerprinting, behaviour models, ban-waves.

Closed loop

Close generate → measure → learn with no human in the publish path, at fleet scale.

Attention priors

Predict which variants will earn attention before a single view is spent.

Self-recovery

Detect persona drift and recover accounts autonomously when performance degrades.

BUILD WITH USHOW IT WORKSSenior engineers + researchers · full-subsystem ownership · meaningful equity. Tell us which problem you’d take.

Infrastructure for the next generation of internet creators.

Project25 is invite-only. If you’re building at the frontier of content and AI, let’s talk.