Simon Späti 🏔️

Fast fuzzy search plugin for large vaults for @obsidian.md. It shows titles first, then headings, then content. Live preview, typo tolerance, finds duplicates, instant on 10k+ notes, uses Smart Connections if present. And much more. Install at community.obsidian.md/plugins/fast....

Also, everything deliberately does not use AI/LLMs, only embeddings through Smart Connections that are created locally. Smart Connection creates local embeddings in `.smart-env`, and we use these to show similar notes. Everything else is just a local index based on your notes.

Are we going back to test-driven development by prompting the tests first and asserting what you want before the agent builds it? But what about the BI engineer? Where does verification happen on your team today? My two cents below.

To me, verification is the liability in agentic engineering today. I explore: - Why "business people are the best data contracts" →Five hierarchy levels of testing across the data stack →Why you can't verify logic spread across four tools and repos Example of five testing levels in one repo:

Are we going back to test-driven development by prompting the tests first and asserting what you want before the agent builds it? But what about the BI engineer? Where does verification happen on your team today? My two cents below.

Sameer, the CEO of Metabase, said: “Business people are the best data contracts, they know when something doesn't make sense, they know the numbers. Plus, they feel it in their subconscious". A lot is changing in BI and data generally.

Fast fuzzy search plugin for large vaults for @obsidian.md. It shows titles first, then headings, then content. Live preview, typo tolerance, finds duplicates, instant on 10k+ notes, uses Smart Connections if present. And much more. Install at community.obsidian.md/plugins/fast....

I am building my dream search for Obsidian based on fzf/Telescope for preview style and fuzzy search, Omnisearch for pre-indexing, and Smart Connections for similar notes/use of embeddings. All based on Rust WASM to be super fast, with no local or online AI involved.

My main goal is to find notes quickly (ms), as I do not use folders, and when I create a new note, to be sure that there isn't the same note with a similar name already existing. This gives me the cockpit view that includes in one search what I used to before: three plugins/separate views.

My GitHub contributions, every year since 2011: 8,204 in total. I've been doing that for a long time, it seems. I totally forgot about 2011, 15 years ago. Apparently it's my bachelor's thesis. Crazy how time flies. And it has certainly increased with AI, but I pushed lots before as well.

How does yours look? Generate it at mo.rgan.co/gh/sspaeti. It's fun to dig up your own history a bit.

New plugin: Manage tags, to-dos, and even tasks, check out my Obsidian Trouble Tag Plugin (updated to `v2.0`. Now you can toggle flags in the note or across the full vault. Mark as done, or show only list callouts, keywords, or actual tasks. Add to Obsidian: community.obsidian.md/plugins/tro...

As well as the beloved Fireplace theme, I uploaded the «Fireplace» theme for both Obsidian and Omarchy.

New plugin: Manage tags, to-dos, and even tasks, check out my Obsidian Trouble Tag Plugin (updated to `v2.0`. Now you can toggle flags in the note or across the full vault. Mark as done, or show only list callouts, keywords, or actual tasks. Add to Obsidian: community.obsidian.md/plugins/tro...

It seems that Amazon has no way of sorting out AI books: Today, I see three book covers already clearly generated by AI. So if the cover is already AI-generated, what about the book? No offense to the authors, but when you spend years writing, why not a nice cover? Just an observation today.

Context can't exist in a vacuum, it must be tangible. For example, we have a physical data model with concrete tables that live in the database at the source, on the other side, we have KPIs and dashboards that show some numbers pulled from that data model.

Some of the presentations I look forward to: →Real-Time State and History with Apache Fluss & StarRocks. →Iceberg REST Catalog, OIDC, Making MV Enterprise-Ready. →Sub-Second Data Architecture. →A Year of Building for the Next Query at Fresha →From Unified Real-Time Analytics to a Governed AI Plane

Context can't exist in a vacuum, it must be tangible. For example, we have a physical data model with concrete tables that live in the database at the source, on the other side, we have KPIs and dashboards that show some numbers pulled from that data model.

We might call the newly ingested information "context", typically not the standard tabular format, but raw text and insights from people and teams, we need a bit of a different approach to read, process, and, maintain. This is what 'context engineering' and all the new information context means.

Context can't exist in a vacuum, it must be tangible. For example, we have a physical data model with concrete tables that live in the database at the source, on the other side, we have KPIs and dashboards that show some numbers pulled from that data model.

Data engineering > AI Engineering. Data wrangling was done by 80% of BI engineers, then 80% of the time by data scientists, and today it's done 80% of the time by AI engineers. It's much more automated, but without clear knowledge or direction for agents, you end up in a mess in no time.

These documentaries are just so encouraging and humbling at the same time, there are people who spend 30 years on an open-source project, making the world and the developer environment a better place, just for personal conviction and having some fun.

Thanks CultRepo for making Netflix documentaries for nerdy developers. And thanks to KDE and the relentless work for making Linux more popular. Also TIL—Qt 1.0 was popularized with KDE, which is what all my widgets are made on my Linux desktop I used today (v6).

This is depressing. Who wants to read an AI-generated book, anyone? Be Tim Ferriss and refuse to let AI write his book, his reason: > [..] if you want to retain a function, a cognitive ability, be very careful about what you outsource.

Tim's reasons elaborated, do you agree?

You ingest data, then you model and transform it, and in the end, serve it to whoever needs to see it. But then, someone asks for a change, everything breaks, and you rebuild. > The tools change, the loop doesn't. This is «A Diary of a Data Engineer» to summarize the state of data engineering.

Using DuckDB as a zero-copy SQL connector is probably one of the most common use cases I use DuckDB for, since the very start. I think it actually might be the default for many, very few store data in DuckDB databases. Early on, it was data lakes, now it might be DuckLakes.

Using DuckDB as a zero-copy SQL connector is probably one of the most common use cases I use DuckDB for, since the very start. I think it actually might be the default for many, very few store data in DuckDB databases. Early on, it was data lakes, now it might be DuckLakes.

Read about the other 4 categories of use cases for DuckDB below.