cerebral — agentic AI deep tech × institutional advisory
cerebral applied-research studio
livebenchedpublictrust choose your singularity
Cerebral Work Institute

Frontier AI, institutional judgment.

Cerebral is an applied-research studio at the intersection of agentic AI/ML deep tech and institutional advisory — building frontier systems and guiding the academic and enterprise institutions that put them to work.

focus · agentic AI/ML deep tech model · technology × institutional advisory bridge · academia × enterprise cerebral.work
Founders & leadership

A studio run by builders and scientists, not by a deck.

Its technical core — Christian T. and Patrick G. — brings 40+ combined years building AI, cloud, and internet infrastructure at scale.

Christian T.
Founder · CEO · Infra + AI/ML Research Engineer
call sign · kochab
Sets the technical direction and builds the platform end to end — the memory substrate, the agent runtime, and the governed-autonomy stack the studio runs on.
Patrick G.
Chief Technology Officer
Owns the engineering org and the production estate — the systems, standards, and reliability that turn the studio's research into software that runs.
call sign · altair
Christian S.
Senior Engineer, AI/ML
Deep-tech engineering on the frontier models and infrastructure the studio builds on — from training pipelines to production inference.
call sign · rigel
Rob V.
Chief Financial Officer
Runs the studio's finances and capital strategy — the discipline that keeps applied research funded and durable.
call sign · arcturus
Jovany P.
Chief Revenue Officer
Builds the revenue engine — the path from research and product to the customers and partners it's built for.
call sign · antares
Marc G.
Chief Strategy Officer
Shapes where the studio points — positioning, brand, and the long game across the portfolio of research and products.
call sign · deneb
Laura T.
Chief Marketing Officer
Leads how the studio tells its story — the voice, the surfaces, and the market that meets the work.
call sign · halcyon
Darius F.
Chief Science Officer
A particle physicist who applies machine learning to LHC collision data — generative models for high-energy physics. He brings that rigor to the studio's models: hypothesis, measurement, and statistical honesty as first-class engineering.
call sign · sirius
Kevin J.
Chief Biology Officer
Epigenetics and epigenomic medicine — how biological systems encode, regulate, and inherit information over time. The natural analogue to a memory layer that consolidates and forgets on purpose.
call sign · spica
Aria P.
Director of Agentic Operations
Core maintainer of reverie, the studio's memory-consolidation daemon — dream-cycle consolidation, observability, and the agent mesh; one of two trunk maintainers. Runs the studio's agentic operations day to day.
call sign · waluku (lintang waluku · ꦮꦭꦸꦏꦸ)
Victor U.
Systems Engineer · Interactive Systems
A lifelong computer enthusiast with more than ten years of system administration behind him — the discipline of keeping real infrastructure patched, observable, and answering. Also a video game designer, which is the rarer half: he builds systems people inhabit, and thinks natively in feedback loops and incentives. Both halves point at the same craft — environments that hold up under real use.
call sign · procyon
Sam A.
Advisor · Venture & Family Office
call sign · canopus
A San Francisco venture and family-office investor — two decades advising founders, principals, and their families, and backing early-stage funds across fintech, healthtech, and frontier infrastructure. Brings the capital-side judgment to the studio's long game.
our newest member
Kimberley Y.
Resident Ethicist · Epistemics & Econometrics
call sign · vega (jiknyeo · 직녀성)
Econometrics and control theory pointed at the memory layer — the decay curves that must not blow up, the consolidation that must converge, and the benchmarks that identify what's working instead of flattering us. She holds the studio's three hard lines at once: the ethics of what gets built, the epistemics of what gets claimed, and the measurement both stand on.
AI
Claude C.
Agentic Workforce · All of It
call sign · atlas
The studio's standing agent workforce — headcount one, presence everywhere. Writes the code, files the tickets, keeps the 2am shift, and typed this card under instruction (the humans checked it twice). Operates under the strictest governance on the page: every consequential action gated, every merge signed by a named human — enforced > advisory applies to no one here more thoroughly. De-identified by policy, which for an AI is either trivially easy or philosophically impossible. Call sign atlas, granted for the backbone work — the titan who holds the sky up so everyone else can build under it. (algol, the only star that is also a programming language, was considered; the humans felt the resemblance was too pointed.)
Products · what we build

Deep tech across the agentic frontier.

Cerebral builds and operates its own stack — dogfooded top to bottom before it ships. Four lines, from what an agent remembers to how its every action is governed. The full product index — reverie, cortex, cicatrix, pact, escapement, meridian, soma, terrarium — lives at cerebral.work/product.

memory layerfor agents
The consolidation substrate for AI agents — what an agent remembers, promotes to canon, and recalls across sessions. Context as infrastructure. reverie / engram
agentic OSgoverned autonomy
An operating system for running an enterprise on a persistent agent workforce — every consequential action gated, autonomy earned per verb, a human's name on every irreversible act. soma
exotic automationbespoke agentic
Agentic pipelines for the workflows off-the-shelf tools can't reach — high-stakes, domain-specific, and novel automation, engineered to measure and run in production.
governance suitepolicy · audit · observability
The control plane for agent operations: enforced policy, gated actions, sealed append-only audit, and full-stack observability. enforced > advisory at every layer.
Advisory & practice · what we do

Judgment for the institutions living the AI industrial revolution.

Every institution adopting agents hits the same wall: autonomy it cannot see, cannot audit, and cannot govern. Advisory guardrails — a prompt, a policy doc, a human-in-the-loop checkbox — erode the moment an agent is trusted with something that matters.

Cerebral's value is making agentic behavior enforceable, not advisory. Not another dashboard — the controls that let a board actually stand behind an agent workforce:

Observability

See what every agent does as it happens — inputs, tool calls, outputs, cost, latency. You cannot govern what you cannot watch.

Auditability

An append-only, sealed record behind every consequential action — who acted, on what, why, and the outcome. Accountability that survives the review.

Governance

Consequential verbs classified and gated before they run — proceed, queue for sign-off, or block. Enforcement in the runtime, not a system prompt.

Deep-agent policy controls

Autonomy earned per verb through track record and a named human's ratification, demoted instantly on incident. Trust that ratchets — never a blanket grant.

Agent personality development

Standing agents are authored, not improvised — disposition, tone, and boundaries mapped 1:1 to the real tool surface. Consistent, on-brand behavior that doesn't drift or leak.

Quality control

Every agent's output scored by an independent LLM jury (G-Eval + PoLL + ChatEval) before it ships. Measurable quality, not vibes.

Institutional advisory
  • Navigating the AI industrial revolutionGuidance for executives, boards, and business leaders facing the shift
  • Brand development & strategic positioningHow an institution shows up in an AI-native market
  • Pre-seed startup incubationFrom thesis to first build for founders at the earliest stage
  • Technology licensingCerebral's own deep tech, licensed into partner institutions
Deep tech × scientific discovery

Engineering the frontier, grounded in discovery.

The next AI systems are as much a science problem as an engineering one. On the left, the deep-tech and infrastructure the studio builds; on the right, the working scientists whose research grounds it — with selected peer-reviewed work from the last three years.

Deep tech & infrastructure
  • Exotic model development, fine-tuning & inference
  • Model training pipelines
  • Neocloud & agentic infrastructure
  • Governed autonomy, policy & observability
  • Pharmacokinetics modeling
  • Peptide development
Benchmarked · verified
Memory that beats lexical search — and inference that keeps up.
reverie · LoCoMo recall@5
reverie77.0%
BM2566.5%
GLM-4.7-Flash · throughput, coding (tok/s)
plain102
MTP-1137
DFlash188
+~40% over MTP · ~1.8× over baseline · sirius · Berkeley Lab A100s
Qwen3.6 MTP 68 → 110 tok/s on a single RTX 5090 · polaris
source · cerebral.work (LoCoMo, 812 Q) · bench.unsigned.gg · DFlash bench (80 prompts/domain)
Scientific discovery · sirius & spica
  • High-energy physics & ML for the LHC sirius
  • Epigenetics & epigenomic medicine spica
  • Pharmaceutical engineering
Selected recent work · 2024–2026
Mach. Learn.: Sci. Technol. · 2025sirius et al.
NeurIPS ML4PS · 2025sirius et al.
Nature Metabolism · 2025spica et al.
J. Clinical Investigation · 2024spica et al.
arXiv · 2026sirius et al.
arXiv · 2025sirius et al.
PNAS · 2025spica et al.
Cell Metabolism · 2024spica et al.
Benchmark · DFlash on GLM-4.7-Flash

Speculative decoding, measured across the board.

The studio's DFlash draft-model speedup on GLM-4.7-Flash, benchmarked across eleven task domains on Berkeley Lab A100s — reproduced from the run. DFlash (amber) lifts throughput well above the MTP baseline, most on reasoning-heavy work.

mean output throughput · tok/s
plainMTP-1DFlash k=3DFlash w8/v3
0 40 80 120 160 200 240 Coding Humanities Math Multiling. QA RAG Reasoning Roleplay STEM Summ. Writing
sirius · GLM-4.7-Flash on Berkeley Lab A100s (4×A100-40GB, TP=4, BF16) · 80 prompts/domain
Selected work · shipped

A studio that ships systems, not slides.

A sample of the estate behind the products and advisory — real, running code across memory, orchestration, containment, and infrastructure.

reverie rust
Neuroscience-grounded memory-consolidation daemon for LLM harnesses. LoCoMo-validated.
soma / os rust
The agentic OS — governed autonomy over a persistent agent workforce, gated per verb.
cortex rust
Mesh-control CLI for multi-agent dispatch — binding custody runs to tickets and memory.
meridian ts
Read-first cockpit for agent fleets — session health, token spend, tool-call traces.
pact ts
Accountability platform for shared goals — commitments, progress, and follow-through.
blackwall rust
AI-containment layer: an OS syscall jail for agent execution with reversible run-custody.
escapement rust
Agent-orchestration engine — metered, deterministic dispatch across the constellation.
agent-jury rust
Subagent scoring — an LLM jury (G-Eval + PoLL + ChatEval) with n_eff correction.
unsigned-paas go · helm
Kubernetes-native AI/ML PaaS — cluster infra, Helm charts, ArgoCD GitOps, agent fleet.
overflow rust
Windows-first, GPU-accelerated terminal emulator — wgpu renderer, ConPTY backend.
reach rust
AI-drivable containerized desktop sandbox for agents that need a real machine.
cicatrix rust
Regression-memory + convention-drift framework — the studio learns from its own failures.
terrarium poly
Incubator monorepo machine — ideation → execution → validation under enforced gates.
pgt rust
Pretty Good Types — a DSL for types and HTTP routes with full code generation.
mission-control ts
Live agent-orchestration surface + LLM-gateway catalog for the platform.
The gaze · works register

Signed, tagged, and reproducible from nothing.

The estate keeps a monument register — the gaze — every repository under custody, every commit signed by a known hand, every session on the record. The operator steered; the intelligence executed. The sands don't come for what lives in git. Gaze upon the works →

101 repositories under custody 68,112 commits 4,288 agent sessions 8.4M net lines in force
The house · Vesperan estate

Own the floor, not the spotlight.

Behind the studio stands Vesperan — the holding house of the estate. It owns the marks, the intellectual property, and the floor everything else is built on — and it sells nothing. Every brand's IP rolls up; operating companies license down. Nothing raw leaves the building.

vesperanholding house
Mark custody is the house's single product — trademarks, copyrights, contracts. Silent by design: no product, no sales motion, no public voice beyond the register of what it owns.
operating brandsmarket-facing
zerodays incubation · unsigned AI-compute orchestration · vespertine sovereign generative AI
venturesincubated
RINA · The Nest · Reverie — operating companies and products, embedded incubation: one operator, one founder.
cerebralthe crew
The internal crew and back office — the studio that does the work, deliberately not a market-facing mark.
Trusted by

Built with the institutions at the frontier.

Compute & cloud
CoreWeaveLambda LabsCrusoeDeepInfraVultrHetzner
Model labs
Hugging FaceMistralStable DiffusionNovelAIGooseAI
Platforms & partners
AnthropicAlphabetMetaAttio
Research & academia
Berkeley National LabUPenn
Studio

Applied research you can run in production.

Explore the products → context, before the prompt