AI Case Study: How do you reduce hallucinations by over 80%? Start with a robust evals framework. A look inside our project teaming up with Zapier on their awesome AI-powered API integration builder: buff.ly/3YzTsg8 Big thanks to @braintrustdata, our go-to evals tool!

The biggest winners in generative AI will be: 1. NVIDIA & TSMC 2. The foundation model company that separates themselves from the pack. Currently this is OpenAI. Time will tell if they keep their lead or if the foundation models become commoditized. 3. Private Equity. The

Private equity is poised to win the gen AI boom...if they don't sleep on AI transformation. Our CEO, Chris Taylor shares a framework for getting started.

AI Case Study: How do you take a stream of unstructured data and turn it into something usable? Build an AI normalization system. Here’s how we partnered with the Sincera team to unlock the value of their data. fractional.ai/case-study/how…

The impact of AI Assist since its September launch has been incredible: more connectors built, more time saved for data engineers. Missed last month’s event where we broke how we built AI Assist with the Airbyte team? engineering.fractional.ai/your-questions…

A lesson from a recent AI project for Zapier : find ways to automate the creation of your 'ground truth dataset' Instead of manually building an eval dataset, we tapped into Zapier’s historical codebase — pairing resolved tickets with merge requests to automate ground truth

5 AI myths keeping companies from getting AI into production...and what to do instead. The latest from our CTO and Co-Founder Eddie Siegel fractional.ai/blog/5-ai-myth…

AI makes mistakes. It hallucinates. It’s not perfect. But that doesn’t mean you can’t trust the systems it powers. Reliable systems can be built with unreliable ingredients. Here’s a practical guide on how: fractional.ai/white-papers/h…

AI Voice Agents: The Good, The Bad, The Ugly Catch our CTO, Eddie Siegel, at the AI Engineer Summit as he breaks down the realities of building an AI voice agent to conduct research interviews Thanks to Latent.Space, swyx, @ai_dot_engineer, and everyone who helped

GPT-5 mini and nano have their own personalities. We tested it with agents solving real problems for enterprises at Fractional AI - some quirks we noticed: * They like complex wording and asking long, multi-part questions * Nano has a tendency to generate would-be penultimate

Even the smaller GPT-5 models are impressive. We dropped 5-mini into a complex “natural language to sql” task we’re building for a pharma services company at Fractional AI. It outperformed o3, o4-mini, 4.1, and 4.1-mini in every one of our evals with no prompt tuning at all.