Wake me up when SAP provides productivity tools of the level of PalantirFoundry’s Pipeline Builder, Ontology Manager, Data Lineage and Workshop.

I'm really glad I don't know what any of those things are.

They are the tools why PalantirFoundry is worth every penny

[not entirely true note: PalantirFoundry is a lowCode/noCode stack on top of a Spark cluster]

Pipeline Builder : https://youtu.be/WIQA7u1tbsY?si=iG-N9xHpUPPVAqDX

OntologyManager: https://youtu.be/SOW0IA_I0bk?si=u5eha0o-yrn4pHvu

Workshop: https://youtu.be/Uh0zpMUR6wY?si=BClTTbch70ZolRCb

I am sorry but these look like very basic ETL/UI Dashboard tools. Is this all they got?

I would love to know an alternative to that superbly integrated stack . [be aware that none of those tools require coding knowledge, nor any DB administration, nor scalability consideration. But for me the most important part is that the data workflow is not dataFrame driven, but semantic entity/relationship driven. Think of that as if you had an ORM magically wrapping your Spark cluster, and make it look like a graph database. Oh, and the sync with ElaticSearch is built-in for a no-brainer semantic search feature on your objects]

> be aware that none of those tools require coding knowledge, nor any DB administration, nor scalability consideration.

These are slogans with marketing grade truth, not engineering truth. The moment you leave the happy path wizards you run straight into Python, SQL, JVM knobs, and cluster limits. The visual Pipeline Builder drops you into a Code node the second you need anything beyond basic joins/filters. Here is an example: https://www.palantir.com/docs/foundry/code-workbook/getting-...

>> nor scalability consideration. Foundry Spark jobs expose tunable variables, you are expected to tweak them for performance and cost: https://www.palantir.com/docs/foundry/optimizing-pipelines/s...

99% of our pipelineBuilders are purely based on the available building blocks [that cover most of PySpark functions: https://www.palantir.com/docs/foundry/pipeline-builder/funct...]. Most notably we do join/filters/explode/group, Dozens of them, realigning data from different parts of the company.

Having most of those PySpark functions exposed visually as building blocks with debug points available at each step simply outperforms a codeWorkbook in term of maintainability and readability.

Yeah as a general rule, when something requires "no coding" I question if they're too zoomed-in. In reality, you can just hire a programmer instead of a box-clicker.

...Except in this case, there might be a lot of value in letting James Bond or Boris Badenov whip up a quick spy pipeline without putting up a job ad for a Python coder.

Once you have used a visual block-based version of PySpark, going back to regular Python code will really look like going back to prehistory.

> I would love to know an alternative to that superbly integrated stack

This one has 10 years of precedence: https://www.ibm.com/products/information-server

> but semantic entity/relationship driven

This is marketing speak who means nothing. What do you mean? Something like this?: https://www.ibm.com/products/information-governance-catalog

Your data in Foundry are available both as tabular content and as a graph at the same time.

Here is a tool to visually explore your graph:

https://www.palantir.com/docs/foundry/vertex/graphs-explore

And here is the search engine to search for your objects (free-text and/or types): https://www.palantir.com/docs/foundry/object-explorer/gettin...

And here is the tool to manage the ontology (the types and relationships of the graph, plus the mapping of the graph to the tabular datasets):

https://www.palantir.com/docs/foundry/ontology-manager/overv...

The UI designer based on the graph data is called Workshop:

https://www.palantir.com/docs/foundry/workshop/overview/

I still have to find a stack where this duality is as integrated as in Foundry.

Ontologies, semantic relationships, data pipelines, data provenance (I suppose a.k.a. lineage), etc. are all things that quickly become familiar to anyone working on knowledge management.

All of these things are pretty fascinating, as they concern the meta of the meta; the nature of knowledge itself; how to slice the ineffable reality into parts and how to communicate complex things to people.

Many software engineers are happy not knowing any of those terms. This perceived boredom is part of the reason why companies, some indeed with questionable ethics, like Palantir (is there even more than one in this space?) have very little competition and are making big bucks on big contracts (and only stand to make more once ML becomes less chaotic and more structured).

I definitely agree.

Two comments here:

- the real power of Foundry is that it is REALLY scalable. Palantir is selling all the knowledge management+data management that you mentionned BUT at a level of integration and scalability that I had never seen in any other product.

- someone in the discussions mentionned IBM InformationServer as an alternative. Functionnaly that sounds quite similar to Foundry. The question is whether the implementation choices are as clever as in Foundry. #toBeInvestigated

> how to slice the ineffable reality into parts and how to communicate complex things to people

So, lots and lots of pie charts?

Slicing reality doesn’t make it into a pie!

You are aware the product names are cognitive obfuscation that the underlying implementation is just vanilla open source data modeling services right?

Linux. Windows. Both are "operating systems"?

What actual novel data collating and anomaly detection algorithms does Palantir implement that no one else does? Oh that's right these services are all just different corporate branding atop the same old open source stacks and libraries everyone uses.

Show me a fully graphical version controlled UI that hides PySpark behind visual widgets.

Show me a entity-relationship model management [with version control] that wraps Spark data frames into a fully ontology-based graph database.

Show me a low code/no code UI designer that is fully consuming the ontology layer abstracting your Spark dataFrames.

#meWaiting

Yeah ok let me go expropriate other peoples systems.

They exist but are used in house not sold as products.

There are information theory texts with all these ideas written in them published back in the 1950s we just didn't have powerful enough computers

Thiel and Musk's only advantage is age. They were able to implement first due to a single property of physics.

Most "geniuses" today, like Tao, just like reading math. None of them have any truly novel net new ideas. They're parrots whose cages were in the nerdy section of a book store.

Pretty sure Tao’s had one or two new ideas.

I still wait for my list of immediately available products, so my company can escape the evilness [and overefficiency] of the PalantirFoundry stack.

This is like the equivalent of "my girlfriend goes to another school, you wouldn't know her". Okay, so there's in-house systems as good as this. Big whoop if no one can actually see or use them.

At least Astronomer has Gwyneth Paltrow

I doubt anyone has the $$$ horsepower to maintain a coherent stack as big as Foundry. So in the makeOrBuy, you probably will go for a buy. And in that field, I still think SAP is way behind Palantir. Someone in the discussion mentionned IBM’s InformationServer, and I will spend some time to investigate its storage scalability strategy and the data model governance.