canopy VS FsCheck

Pinecone: A scalable vector database service that facilitates efficient similarity search in high-dimensional spaces. Ideal for building real-time applications in AI, such as personalized recommendation engines and content-based retrieval systems.

In this post we’ll build a Python script that uses StripeDocs Reader, a loader on LlamaIndex, that creates vector embeddings of Stripe's documentation in Pinecone. This allows a user to ask questions about Stripe Docs to an LLM, in this case OpenAI, and receive a generated response.

Pinecone is a managed vector database service that simplifies the process of building and scaling vector search applications. It offers a simple API for embedding vector search into applications, providing accurate, scalable similarity search with minimal setup and maintenance.

In case of AIMD, I am doing this all in-memory, but you could also do this in a database (e.g. Pinecone). It all depends on how much data you have and how much compute you have available.

A RAG implementation's quality and performance highly depend on the similarity-based search of embeddings. The challenge arises from the fact that embeddings are usually high-dimensional vectors, and the knowledge base may have many documents. It's not surprising that the popularity of LLM catalyzed the development of specialized vector databases like Pinecone and Weaviate. However, SQL databases are also evolving to meet the new challenge.

In the dynamic world of web development, Python has emerged as a dominant force, especially in backend development – the primary focus of this blog post. Although it's worth mentioning that there are ongoing efforts to use Python for the frontend as well, like Reflex (previously known as Pynecone, they presumably had to change their name because of Pinecone vector database), which even garnered support from Y Combinator. Samuel Colvin (creator of Pydantic) is also working on FastUI (he literally just released the first version in December 2023).

As you can see from the imports statement we're relying on FsCheck to generate some random values for us.

C# - FsCheck

https://fscheck.github.io/FsCheck/ can hopefully generate random inputs automatically or with low effort for many methods to get your code coverage up. You don’t even need to write real tests right now, just call the methods with the random inputs and check they don’t fail.

Try using F# for tests. It has some great libraries like FsCheck (https://fscheck.github.io/FsCheck/).

At Symbolica we're building a symbolic execution service that explores every reachable state of a user's program and verifies assertions at each of these states to check that the program is correct. By default it will check for common undefined behaviours, such as out-of-bounds memory reads or divide by zero, but it can also be used with custom, application specific, assertions too just like the kind you'd write in a unit test. Seen from this perspective it's kind of like FsCheck (or Haskell's QuickCheck or Python's Hypothesis), but much more exhaustive and without the randomness.

Maybe something like QuickCheck, a quick search gave me this library for .NET https://github.com/fscheck/FsCheck

Because of the above I've generally been using tools like Stryker.NET and FsCheck to augment my testing suite. I'm still doing unit testing to find the more obvious "I haven't had my coffee, let's make sure I'm doing what I think I'm doing" bugs. I'm just using things like mutation testing, property testing, fuzzing, etc. to find the deeper issues in my code. There's a ton of libraries out there, including one that I've built for myself to help with testing but FsCheck and Stryker are just beautiful. And if you're interested in fuzzing, SharpFuzz is a great option. But that one isn't quite as easy of an on ramp compared to the other two that I mentioned.

Looks cool. Is there a reason why you didn't use FsCheck or Hedgehog? They're built to generate random data for testing, and can return the seed if a test fails so you can rerun the test with the exact same data once you figure out what the problem is - which is useful if the failure condition is rare.

Haskell has QuickCheck and Hedgehog, and dotnet has both as well. F# is favored, but there's C# interop.

Side note, if you are thinking about testing in general, might be interested in property based testing. See for example https://fscheck.github.io/FsCheck/