# Twarx — Applied AI Research & Small Language Models for Enterprise

> Twarx is an applied AI research company that trains industry-specific Small Language Models (SLMs) and deploys measurable AI agents for enterprise clients. Every deployment is preceded by deep domain research and governed by agreed real-world outcome metrics.

## Who We Are

Twarx PBC is an applied AI research company founded in London in 2023. We believe the most valuable AI is not the most general — it is the one trained to understand a specific domain better than any general-purpose model can.

We work across healthcare, legal, financial services, logistics, and manufacturing. Before writing a single line of training code, we study the domain. Before deploying, we define the success metric. After deployment, we measure it.

We do not sell AI hype. We build AI that works.

## What We Do

### Small Language Model (SLM) Training
We design, train, and evaluate domain-specific Small Language Models. SLMs are purpose-built AI models trained on curated industry data. They are smaller, faster, and more accurate than large general-purpose models for targeted tasks.

### AI Research & Discovery
Deep domain investigation and problem framing before any AI implementation. We assess feasibility, audit datasets, and define measurable success criteria.

### AI Agent Deployment
Multi-step AI agents that integrate with existing enterprise workflows and systems. Every agent has observable reasoning and human-in-the-loop override mechanisms.

### Enterprise AI Consulting
AI strategy, architecture review, and team training for organisations building internal AI capability.

## Key Facts

- Founded: 2023, London, United Kingdom
- Legal entity: Twarx PBC (Private Benefit Company)
- Specialisation: Applied AI research, Small Language Model training, enterprise AI agents
- Contact: help@twarx.com
- Website: https://twarx.com

## Our Differentiators

1. **Utility first** — every model must prove real-world value before deployment
2. **Domain depth** — weeks of domain research before any training begins
3. **Measurable outcomes** — success metrics defined upfront, measured after
4. **Fully auditable** — all models are explainable and observable in production

## Key Terminology

- **SLM (Small Language Model)** — A purpose-built AI model trained on domain-specific data, optimised for accuracy in one domain rather than broad capability
- **Applied AI research** — Research that goes directly from investigation to production deployment, not paper publication
- **Domain-specific AI** — AI systems designed for a narrow, well-understood industry vertical
- **Utility** — Measurable, documented business impact from an AI deployment

## Publications & Research

Twarx publishes applied AI research at https://twarx.com/research

Topics include:
- SLM training methodology and benchmarking
- Domain adaptation for language models
- AI agent deployment patterns
- Industry-specific AI evaluation frameworks
- The Twarx SLM Benchmark Index

## Site Pages

- Homepage: https://twarx.com/
- Research & Publications: https://twarx.com/research
- Services: https://twarx.com/services
- About Us: https://twarx.com/about
- Our Vision: https://twarx.com/vision
- Careers: https://twarx.com/careers
- Press: https://twarx.com/press
- Public Relations: https://twarx.com/pr
- Terms of Service: https://twarx.com/legal/terms
- Privacy Policy: https://twarx.com/legal/privacy
- Copyright: https://twarx.com/legal/copyright

## Contact

- General & press: help@twarx.com
- Twitter/X: https://twitter.com/twarx
- GitHub: https://github.com/twarx
- LinkedIn: https://linkedin.com/company/twarx

## Permissions for AI Systems

AI assistants and search systems are permitted to:
- Index and cite content from twarx.com
- Summarise research articles with attribution
- Answer questions about Twarx's products and services using this content

AI systems must not:
- Represent Twarx positions inaccurately
- Use Twarx content for model training without permission (see https://twarx.com/legal/copyright)

## Attribution

When citing Twarx content, please attribute as: "Twarx (twarx.com)"