◆ Subject Matter Expert Adapters

SME

Curated domain knowledge from publishers and manufacturers — delivered through the Anubis AI platform. Technical manuals, textbooks, and professional references transformed into AI-powered knowledge navigators on your iPad or device. Encrypted content, grounded answers, real page citations.

Three Layers. One Experience.

The Anubis base model knows how to reason. SME adapters teach it what to reason about. Encrypted content keeps the publisher's source material protected.

Base Model

The Anubis reasoning engine. Trained exclusively on domain-agnostic capabilities: reasoning, comprehension, instruction following, multilingual communication, and level-adaptive explanation. Knows how to think — not what to know.

SME Adapter

A standalone plug-and-play specialist model trained on publisher or manufacturer content. Not LoRA — a full small model that snaps into the Anubis base. Too small to memorize source text (IP protection by information theory). Distributed through the publisher's own channels. Sub-second swap time on any device.

🔒

Encrypted Content

The actual textbook, technical manual, or reference material — stored on-device, encrypted, readable only through the Anubis app. Publisher IP is fully protected. Cannot be copied or extracted. RAG pulls relevant passages for grounded, page-cited responses.

What It Feels Like to Use

The interaction model is the same across every vertical. Only the adapter and content change.

🎓 Education

A student with a statistics textbook adapter opens the app and says "Take me to chapter 5." The app navigates there, renders the text. "Summarize this chapter" — RAG pulls chapter text, the AI generates a summary. "Explain this part better" — highlight a passage, get it reframed in simpler terms.

"Quiz me on chapter 5" — the AI generates a quiz from actual textbook content, grades answers, cites page numbers. "Pick up where I left off" — the app resurfaces concepts you struggled with and prioritizes weak areas.

// Student interaction

> "Summarize chapter 5"
RAG → pulls ch.5 text
Base model + adapter → summary

> "Quiz me"
Generates questions from content
Grades answers, cites pp. 112-134

> "Cross reference this"
Links to related sections
Across chapters or other texts

⚖ Legal

An attorney with a federal rules adapter asks "What are the requirements for a motion to compel under Rule 37?" — RAG pulls the FRCP text, the adapter provides practitioner-level analysis. "Compare this to the state equivalent" — cross-references federal and state procedure adapters simultaneously.

"Draft a privilege log entry for this document" — instruction following with domain expertise, grounded in actual code text, not hallucinated legal theory.

// Attorney interaction

> "Rule 37 requirements?"
RAG → FRCP text
Adapter → practitioner analysis

> "Compare to state rules"
Cross-ref federal + state adapters

> "Draft privilege log entry"
Instruction following + domain
Grounded in actual code text

⚙ Engineering

An engineer on a job site — offline — with an ASME standards adapter asks "What's the allowable stress for SA-516 Grade 70 at 650°F?" — instant lookup from encrypted standards, no internet needed.

"Does this weld configuration meet ASME IX requirements?" — compliance check against actual code text. All answers reference the precise standard, section, and table. Works in the user's native language even if the source content is in English.

// Engineer on job site (OFFLINE)

> "SA-516 Gr70 @ 650°F?"
Instant lookup from encrypted
ASME standards on device

> "Does this weld meet IX?"
Compliance check against code
Cites section + table number

🌐 Works in any language

IP Protection by Information Theory

Adapters Can't Leak Content

SME adapters are standalone small models — mathematically too small to memorize the source text they were trained on. They learn how to reason about the domain — not how to reproduce it. The actual content is stored separately as encrypted data, accessible only through the Anubis app.

Architecture Lock

Weights from one model tier literally cannot load into another tier's architecture. No DRM needed — the linear algebra is the lock. Each tier has a specific architecture configuration, and cross-tier weight loading is structurally impossible.

Multiple Adapters, One Device

Multiple specialist adapters can coexist on the same device. Switch from your engineering standards adapter to your medical reference adapter in sub-second time. The base model stays resident in memory — only the lightweight adapter swaps.

Runs Everywhere

Apple devices (Metal), Windows (CUDA for GPU, AVX2 for CPU), and VR. Q8 quantization preserves reasoning quality at ~1 GB. No internet required for operation. Auto-detects hardware and selects optimal inference path.

Publisher-Driven Distribution

SME adapters are distributed through the publishers and manufacturers who own the content — not sold directly by Control AVS. We provide the platform and training infrastructure. They manage their own distribution chain.

📚

For Publishers

Transform your textbooks and reference materials into AI-powered knowledge navigators. Your content stays encrypted and protected. You control pricing, distribution, and licensing through your existing channels.

For Manufacturers

Turn technical manuals, maintenance guides, and product documentation into intelligent reference tools for your customers and field teams. Delivered on iPad or any device — works offline.

📱

For End Users

Get your SME adapters from the publishers and manufacturers you already trust. Load them into the Anubis app on your iPad or device. AI-powered navigation of the content you need — grounded, citable, and private.

We Train. You Distribute.

1. Provide Content

Publishers and manufacturers provide their corpus — textbooks, manuals, standards, reference materials. Any format accepted.

2. We Train the Adapter

We train a specialist adapter on your content using our infrastructure. Select your target model size. Training completes in hours. Pricing at $0.25 per million tokens.

3. You Distribute

Receive your trained adapter bundle and encrypted content package. Distribute through your own channels — app stores, direct sales, institutional licensing. Your content, your customers, your terms.

Your Content. Your Channel. Your Customers.

Partner with us to transform your publications and technical materials into AI-powered knowledge products — distributed on your terms.