AI codegen feels magical, until you try to build on top of it. It’s not that the code is broken. It just doesn’t fit. The abstractions are off. The state is scattered. You end up rewriting it all. AI needs opinionated defaults.
Read more: www.preswald.com/blog/opensrc...
Structure that makes the right thing easy and the wrong thing hard. That’s what Preswald gives you: deterministic execution, reactive components, and a clear mental model for building data apps.
And open source matters more than ever. Because the scaffolding you build today becomes training data
🚀Preswald OSS Launch Week Day 1: API data sources
Connect any REST API from config
→ response is parsed + loaded into DuckDB
→ query it like a database
Thanks to Lokesh Velayudha (github.com/lokeshvelayu...) for the contribution!
github.com/StructuredLa...
Preswald just hit 2K GitHub stars. Onward and upward!
github.com/StructuredLa...
We put together a step-by-step guide on using Preswald to turn raw chatbot logs into an interactive debugging dashboard.
www.preswald.com/blog/ai-chat...
Pyodide lets you run Python right in the browser. No installs, no setup, just open a link and go. It makes sharing and running data apps so much easier. Try it out with Preswald today: github.com/StructuredLa...
CSVs Are Not Databases: The challenges of local data exploration
open.substack.com/pub/structur...
🦆MotherDuck → DuckDB in the cloud + read scaling for blazing-fast queries
📊 Preswald → Declarative, Python-native dashboards with ultimate interactivity
Blog: motherduck.com/blog/preswal...
Code: github.com/StructuredLa...
@Structured teamed up with @MotherDuck to show how you can analyze massive public health datasets 4x faster and turn them into live, interactive data apps in just a few lines of Python, using our OSS framework Preswald.
But the point is, when you're in a flow state, it’s actually physically painful to stop working. Take breaks when you need to, not because of the time on a clock
Reason I don’t believe in the Mon-Fri, 9-5 life: I’ll often have bursts of inspiration/creativity that sometimes last 4-5 days, where I can’t sleep, can’t eat. I have to build. Then, ofc I’ll crash for 24 hrs after.
B2A: The future of analytics isn’t human
For the agents. By the agents.
Read more: blog.structuredlabs.com/p/b2a-the-fu...
Tools created for a specific need, used for a short time, then discarded. Instead of forcing rigid, bloated SaaS into every workflow, individuals can now spin up lightweight, hyper-personalized apps that do exactly what’s needed—nothing more, nothing less.
We’re used to thinking of software as something permanent: bought, maintained, and scaled. But what if it didn’t have to be? The cost of writing code has plummeted, making it easier than ever to build ephemeral software
open.substack.com/pub/structur...
Enjoying a no-meeting Sunday to build. Lots of exciting stuff shipping soon :D
This post walks through setting up a real-time AI debugging dashboard to spot ASR errors, intent misclassifications, and slow responses, so you can fix issues fast and improve performance.
Debugging AI voice agents is way easier when you have the right dashboards. With DuckDB & Structured, you can pull logs locally, run instant SQL queries, and turn messy data into clear, interactive dashboards.
blog.structuredlabs.com/p/debugging-...
Empirical results demonstrate that DeepSeek-R1 significantly outperforms OpenAI-GPT, achieving a 27% higher groan-to-laugh ratio and a 42% increase in subjective user satisfaction.
, humor remains an underexplored yet impactful dimension of human-AI interaction. Through a rigorous A/B testing framework, we assess the models on joke originality, coherence, and laugh-induction rates.
We present a comparative evaluation of two state-of-the-art language models, DeepSeek-R1 and OpenAI-GPT, focusing on their generative performance in a specialized domain: knock-knock joke creation. While traditional benchmarks evaluate language models on metrics such as perplexity or BLEU scores
Excited to speak at the AITinkerers San Francisco meetup on Jan 29! I'll be showing how to set up lightweight local analytics dashboards w/ Python. Don't miss it!
sf.aitinkerers.org/p/ai-tinkere...
Structured blog is now officially live: blog.structuredlabs.com
We’re sharing technical deep dives, real-world challenges, and how we’re building Structured to solve complex AI and analytics problems.
If that’s your thing, check it out and subscribe today. 👉
By acting as a middle layer, Ibis addresses challenges like fragmented storage, scalability, and redundant logic, enabling a more consistent and efficient approach to multi-backend data workflows.
It provides a DataFrame-like API, similar to Pandas, but translates Python operations into backend-specific queries. This allows it to work with SQL databases, analytical engines like BigQuery and DuckDB, and even in-memory tools like Pandas.
We recently explored Ibis, a Python library designed to simplify working with data across multiple storage systems and processing engines. Wrote up some learnings here: open.substack.com/pub/structur...
They don’t dictate the narrative but amplify human creativity, allowing analysts to focus on interpreting results rather than wrestling with the system.
Need to design a dashboard or explore a new KPI? Analytics agents can suggest metrics, outline relevant dimensions, and even highlight tradeoffs between different approaches.
They collaborate in exploratory data analysis, brainstorming new angles, surfacing unexpected insights, and adapting dynamically as business needs change.
These agents act as partners. They contextualize data, guiding you through what you have and how to use it effectively.
