conradbez

As a data engineer, we finally reached decent tooling for the data stack, but now it feels like we are starting over with ad-hoc prompt pipelines. I built prompt-build-tool to bridge that gap.

The tool treats prompts as templatable code assets rather than static strings. It applies pipeline stability and data quality lessons from DE to the LLM lifecycle:

Declarative Pipelines: Build workflows by referencing outputs of previous prompts.

Granular Testing: Test prompts "segments" at the smallest level to quickly find regressions and understand system-level tradeoffs.

Scaling to Production: Manage complex meta-prompts and parallel execution paths while iterating quickly between dev and prod.

https://github.com/conradbez/prompt-build-tool

Inspired by dbt (data-build-tool).

Thanks for checking out

Couple tips on audio front:

1. gemini has native audio understanding so I would recommend checking out uploading there and playing with the prompt to get it's output matching what you are after

2. for audio over 1-hour I found chucking it into 45min segments made it easier for Gemini to give back reliable timestamps

3. you do need to check the LLM outputs for valid timestamps - it can go off the rails

I'll add search with the existing vector embeddings used for recommendation system and audio waves to the feature list - great idea!

YouTube takes free uploaded podcasts clips and charges us outrageous premium fees to view.

The obvious alternative is to use a podcast app but I like the 10-20min clips and recommendation engine on YouTube. So I built podtoc.com to serve LLM generated podcast clips youtube-feed style.

The Tech:

- LLM Pipeline: I built a pipeline to extract meaningful clips from long-form content, specifically designed to capture one of the core insights covered in the podcast.

- Recommendation Engine: Suggests clips based on previous listening to solve the discovery problem.

- The App: A React Native (Web/iOS) app featuring a "swipe to next" UI for seamless browsing.

If this sounds like a problem you’ve faced, I’d love to hear:

1. Which podcasts would you like to see added to the library?

2. Any feedback on the UI or bugs you encounter?

3. Any questions about the pipeline or suggestions for the recommendation logic? (would love to open source after some cleanup)

Check it out: https://podtoc.com/app

Hey guys, at Dispensed are working on Vending as a Service.

For $100 we ship a vending mechanism for you to put on your own custom vending box.

We then provide online marketplace and payment services (Shopify for vending machines) so your customers can use their phones to buy products from your vending machines.

Great alternative for products that don’t justify big $10k vending machines.

http://about.dispensed.app

Very nice - I've recently launched something similar that tracks calories consumed (with macros coming soon). Would love to compare notes sometime if you're keen :)

https://news.ycombinator.com/item?id=40316291

I fully agree that anything to the extreme can be detrimental and therefor unsustainable and ineffective. I will say that I have seen great value in building up a rough mental model of (1) how energy dense various foods are (2) where my routines / habits are causing me to over consume energy.

I agree having a compulsion to weigh everything you eat for fear of "failing" isn't very productive. But on the other hand I think there is genuine utility in knowing roughly what percentage of your daily energy comes from your typical breakfast / lunch / drinks / dinner / snacks etc etc.

Where I feel my approach beats the typical calorie tracker is that it doesn't encourage micro-counting or even reaching a specific goal, but rather just keeps you informed. This makes it very easy to know where your energy came from and thus gives you power to increase / decrease as needed.

Definitely take your point on the other factors - but for many people, I think, it would help knowing what a maintenance day of eating looks / feels like.

Write re-usable sql code blocks and compile it to a dbt project with a yaml defintion

Great for:

- Automating dbt code generation, so you can manipulate yaml instead of sql

- Construct pipeline without knowing SQL

- Construct pipelines without learning dbt

- Avoid duplicative code

Thanks! Yeah interesting idea on the swappable backend, something I'll consider.

Wow this looks awesome! Definitely have a bit of a browse, thanks!

When dash came out I was a massive fan but found the abstraction layer not quite right.

When writing a python script you tend to think from top to bottom - as your script runs through. Dash adopts the html approach of defining a document and callbacks. After doing the refactor to this model I found the trade-offs of Dash wasn't worth the time saved of going straight to a more traditional stack, and again you're pretty locked-in to this choice now.

One example was writing a Dash SSO component turned into quite an ordeal (more so than most SSO integrations) - my aim here is to make an extension like this easier because you've got access directly to the FastAPI app with `hs.app`.

Haven't worked on this in a while so can't exactly remember but focuses more on the MVP functionality. Should be easy enough to add anything you could traditionally do through SSHing into the box though.

Looks very interesting, will have to give it a shot!

I made an open source “dashboard” for dokku that tries to give you Heroku ease of use with the cost of a single server.

Basically you run one script on the server and it deploys a dokku app which manages the deployment of additional dokku apps. Gives you GUI access to deploying new apps, changing env variables etc.

Would love feedback from anyone looking for an easy way to deploy dokku apps regularly.

https://github.com/conradbez/dokku-dashboard