Meta Model API

LLAMA 4 SCOUT AND MAVERICK

70%

reduction in manual compliance reviews

5x

improvement in latency using Llama for on-device inference

60%

more cost-effective in production using Llama compared to other foundation models

^*All results are self-reported and not identifiably repeatable. Generally expected individual results will differ.

CASE STUDY

Delivering a solution for trustworthy, traceable AI to regulated industries

At a glance

Industry

Technology

Use case

Making AI compliance for domain-specific applications more efficient

Goal

Reduce barriers to AI innovation for companies in regulated industries

Llama versions

Llama 4 Scout, Llama 4 Maverick, Llama 3.3 70B Instruct, Llama 3.3 8B Instruct, Llama 3 8B, Llama 3 13B

Deployment

Edge hardware, Amazon Web Services

THEIR STORY

Improving AI trust across industries

Authorship is on a mission to help companies — from law firms to art galleries — safely adopt domain-specific AI tools in full compliance with evolving global regulations. Through private AI infrastructure, they provide end-to-end data governance solutions, spanning from on-device to secure cloud solutions, that are auditable and certifiable and deliver total control.

THEIR GOAL

Solve the trust gap in AI applications

Authorship knew the barrier to AI adoption in regulated industries wasn’t technology, it was oversight. Law firms wanted to create practice-specific models. Art galleries wanted to use AI to verify provenance. Government officials wanted to use AI to improve efficiency. But they all needed to trust that AI models and applications could be trained, deployed and used without breaching compliance protocols immediately and over time.The Authorship team set out to create a certifiable, private AI infrastructure solution that was efficient and scalable. The solution needed to ensure three things: 1) that every model was fine-tuned on domain-specific data with compliance top of mind, 2) every output was traceable, and 3) each deployment — including constrained environments at the edge — was governed by real-world documentation like waivers and privacy agreements.

THEIR SOLUTION

Unlocking secure AI: Authorship’s Llama 4 breakthrough

Authorship used Llama 4 Scout and Llama 4 Maverick — the first multimodal sparse mixture-of-experts (MoE) Llama models — to create a scalable, end-to-end solution that brings control to the AI supply chain. Authorship’s secure AI cloud solution allows organizations to develop domain-specific applications while keeping training data compliant, monitoring inference and ensuring IP protection and transparency. For edge use cases, distilled and air-gapped Llama models give users full on-device control.Using Llama models with sparse MoE architecture reduced compute costs and sped up training and inference because each input is routed to a subset of experts for processing. Llama models also offered a blend of openness for fast onboarding and full data control, competitive performance and deployment flexibility with smaller options. Across industries, companies are already taking advantage of Authorship’s solution:

THEIR APPROACH

Bringing trust to fine-tuning domain-specific models

Authorship balances data privacy, compliance and model performance, delivering a robust solution for domain-specific applications.Stage 1: Data curation and fine-tuning

On-device data curation: A small Llama model analyzes the full training corpus locally on the device, identifying representative examples and excluding sensitive information. Each data point is assigned compliance tags, such as “biometric identifier” or “attorney-client privileged,” and is linked to legal documentation, ensuring a transparent compliance record.

THEIR SUCCESS

Unleashing secure and compliant AI innovation for all

With Authorship’s Llama-backed AI infrastructure solution, organizations in law, athletics, fine art, healthcare, government and beyond can responsibly adopt domain-specific AI solutions and keep control of their data. The solution ensures AI applications are private and auditable and supports efficient, on-device AI deployments such as laptops, tablets and even sporting equipment. That includes constrained environments with no access to the internet. The solution enhances confidence in regulatory compliance and significantly reduces the time and effort required for related tasks.

70% reduction in manual compliance reviews

“Llama offered the best combination of openness, technical performance and deployment flexibility, especially for on-device use and offline inference. It directly supports our goal to make AI systems provably private, auditable and compliant without compromising user control.”

Nils Tracy, Founder and CEO, Authorship

Previous slide

Next slide

Slide 1 of 3

Start building

Download models View documentation

Explore more

ConsumerBusiness

Shopify | Llama case studiesShopify uses Llama to generate product pages, localize content, and automate support, helping developers scale workflows and save time.Read more

ConsumerTech

Scribd, INC | Llama case studiesDelivering faster, cheaper and more accurate results with a Llama-powered AI content discovery assistant.Read more

Tech

Exati | Llama case studiesTransforming support for smart city platform clientsRead more

COA Registry relies on Authorship to provide art galleries with financial data protection, provenance records and certificates of authenticity.
The Cycle Path builds Authorship’s on-device Llama models into their bicycles to support custom, specialized fits with personalized data controls for private health and performance data — like nutrition and aerodynamics — and in-the-moment coaching.
Dynamic employs Authorship’s solution to monitor AI regulations and ensure compliance across jurisdictions while handling wallet infrastructure for crypto and fintech.

Llama models offer a scalable, trustworthy foundation for Authorship’s solution.

Fine-tuning: The curated dataset is then used to fine-tune Llama models like Llama 4 Scout, Llama 4 Maverick or Llama 3.3 70B, resulting in a domain-specific model.

Reasoning template creation: Authorship constructs a teacher-student loop where larger models produce generalized reasoning templates that can be used for smaller, distilled models.

Stage 2: Deployment

The fine-tuned Llama models are deployed in one of two ways:

On-device inference: Distilled models are deployed directly on the device for local inference.
Cloud-based orchestration: Fine-tuned models are containerized and deployed on Amazon Elastic Kubernetes Service (EKS) for cloud-based inference.

Tech stack

Authorship’s solution integrates a range of tools, including:

Pinecone and Weaviate for vector storage
BGE and custom embeddings for region-specific context
LangChain for orchestration
A proprietary compliance rules engine to ensure regulatory compliance

Authorship’s Llama deployment pipeline features a compliance-first architecture.

5x improvement in latency using Llama for on-device inference

60% more cost-effective in production with Llama compared to other foundation models