Research | WRITER

Advancing AI for the enterprise

At WRITER, we have one goal: to build scalable, reliable,

and transparent AI technology for the enterprise.

Our approach

Our approach is different: we believe that building large language models (LLMs) informed by enterprise requirements leads to AI systems that are more reliable, more controllable, and more transparent. When you ground cutting-edge AI innovation in real-life needs, it yields solutions that solve problems people actually face.

Our team

Our globally distributed team of AI/ML researchers and engineers has a five-year track record of groundbreaking research and development across language models, retrieval systems, and evaluations.

Our results

Our results demonstrate that when AI research starts with real needs, it leads to:

Research highlights

Read the blog

Watch the video

Read the blog

Watch the talk

Watch the video

Research pillars

Enterprise-optimized models

Focus on developing more scalable, reliable, and transparent models specifically engineered for enterprise requirements

Practical evaluations

Development of model evaluation methodology that reflects real-world scenarios and risks

Domain-specific specialization

Research into applying AI systems in high-stakes industries

Retrieval & knowledge integration

Work on next-generation retrieval systems that safely and reliably connect language models with enterprise data

All Practical evaluations Enterprise-optimized models Domain-specific specialization Retrieval

How personalized context quietly degrades AI accuracy: a deeper look
Practical evaluations
Jul 26, 2026

Accurate Failure Prediction in Agents Does Not Imply Effective Failure Prevention
Practical evaluations
Feb 3, 2026

Towards Outcome-Oriented, Task-Agnostic Evaluation of AI Agents
Practical evaluations
Nov 11, 2025

Palmyra-mini: Small models, big throughput, powerful reasoning
Enterprise-optimized models
Sep 19, 2025

Reflect, retry, reward: Self-improving LLMs via reinforcement learning
Practical evaluations
Jun 12, 2025

Palmyra X5: The end of context constraints
Enterprise-optimized models
Apr 28, 2025

Expecting the unexpected: FailSafeQA Benchmark
Practical evaluations,
Domain-specific specialization
Feb 10, 2025

Palmyra Creative: Unlocking creativity with AI
Domain-specific specialization
Dec 17, 2024

Introducing Self-evolving models
Enterprise-optimized models
Nov 20, 2024

Palmyra X4: Introducing actions
Enterprise-optimized models
Oct 9, 2024

Writing in the Margins
Enterprise-optimized models
Aug 27, 2024

Comparative analysis of retrieval systems in the real-world
Retrieval
May 3, 2024

OmniACT: A benchmark for enabling multimodal generalist autonomous agents
Practical evaluations
Feb 27, 2024

Fusion-in-decoder: achieving state-of-the-art open-domain QA performance
Enterprise-optimized models
Sep 13, 2023

Becoming self-instruct: Introducing early stopping criteria for minimal instruct tuning
Enterprise-optimized models
Jul 5, 2023

Palmyra Med: Instruction-based fine-tuning of LLMs enhancing medical domain performance
Domain-specific specialization
Jul 3, 2023

Palmyra Fin
Domain-specific specialization
Jul 3, 2023

Grammatical error correction: a survey of the state of the art
Enterprise-optimized models
Apr 29, 2023