Brandon T. Willard

Structured generation is easy to demo, hard to get right. Most libraries silently drop constraints they can’t handle. Your tests pass, then production breaks. We don’t. Read the full blog post by the .txt team here 👇 blog.dottxt.ai/dotjson-has-go…

People falsely believe that structured generation slows down inference... It actually speeds it up! 🏎️ On the left, standard generation. On the right, structured generation with coalescence.

@AsfiShaheen data-monger.com/syndeblog/the-…

The healthcare industry has a data standard so complex it breaks conventional LLM structured generation. We just became the first to generate valid FHIR outputs with Qwen3-4B-Instruct. 👇 A breakthrough in healthcare AI.

Rémi @remilouf presenting Function Calling that doesn’t suck at @aiDotEngineer #aieparis

Happy to support @jsonschema, the standard at the heart of structured outputs.

When Tensorlake works with @dottxtai, you get structured document parsing that improves constrained LLM generation. Clean fragments + schema enforcement = reliable extraction pipelines that don't break on malformed outputs. Blog + Colab Notebook in the comments 👇

The family is growing. More details soon.

We built what OpenAI didn't. Optional fields. Unions. Dynamic keys. Composition. Features that matter when you deploy. 🧵

We just launched dotjson on AWS Marketplace — constrain outputs to JSON Schema without blocking the model’s reasoning. Think first, structure after. Try it here 👉 buff.ly/QOdHO9O

The .txt team is growing. We're looking for a: - Founding DevRel Engineer, - Member of Technical Staff - LLM Inference, and - a Rust Developer Apply here: jobs.ashbyhq.com/dottxt

Edge inference demands blazing-fast structured generation. Our technology delivers exactly that and with grammars we can eliminate JSON overhead for even faster function calling. Excited to announce our collaboration with @liquidai bringing this to edge devices!

We’ve rebuilt dottxt.ai. See what’s new on our site.

Coming soon

During our internal hackathon, one of the teams ("unladen swallow") used structured generation to make our products dance with animated SVG. Enjoy ☀️

🚨 Outlines v1.0 is out 🚨 🌱 Simplified - Use any Python type to define the structure 🧨 More powerful - Expressive language to define complex structure and a library of built-in types. 🚀 Production-ready - Integration with inference servers (vLLM, Ollama, etc.) and APIs

Introducing Snap-Serve! Positron (LLM Inference Systems) + Parasail (AI Deployment Network) Instant Access to Finetuned LLM Models Running on Unbounded Inference Hardware This partnership blends Positron's industry-leading transformer inference servers and Parasail's advanced infrastructure management capabilities to accelerate AI adoption for developers and AI Engineers who require a cost-effective way to host AI agents and personalized LLMs. Parasail's AI Deployment Network gives AI companies fast, scalable, and cost-efficient access to compute and the latest cutting-edge AI models. Offering the industry's largest on-demand GPU pool, Parasail instantly scales from zero to enterprise-level workloads, ensuring guaranteed availability. Teams migrating from proprietary models to Parasail reduce costs by 15-30x with an additional 2-5x cost advantage over other open-source providers through intelligent workload orchestration. Trusted by AI startups and enterprises alike, Parasail delivers unmatched flexibility, control, and performance. 🔗 in comments👇

Check out this cool project by @aastroza to automatically convert any TV stream into a real-time movie script! Amazingly cool project using Outlines to enforce movie script formatting, and generally an impressive project. Here's the entire JSON schema he uses. Repo below

During our internal hackathon, one of our teams (named "Too Many Cooks") wrote a grammar-powered recipe generator. The language model can ONLY generate text consistent with this recipe format. Here's a recipe for the universe.